Decision-making method for optimization design of non-uniform indoor environment in large-space building and based on multimodal-knowledge-graph-enhanced foundation model

Abstract

Provided in the present invention is a decision-making method for optimization design of a non-uniform indoor environment in a large-space building and based on a multimodal-knowledge-graph-enhanced foundation model. During a design decision-making process, a green-performance optimization design model adapted to a typical engineering scenario of a non-uniform indoor environment in a large-space building is constructed, an in-depth understanding of the optimization design problem of the non-uniform indoor environment is gained, and a decision-making method for optimization design is provided. Values such as building morphological parameters, spatial parameters, and heating, ventilation and air conditioning system parameters, green-performance image data of a typical section of the large-space building, etc. are outputted. On this basis, a method for constructing a knowledge-representation-enabled enhanced foundation model is provided. On the basis of an end-cloud collaborative distributed operating architecture, optimal allocation of computing power resources is realized, and the collaboration of a multimodal knowledge graph and a vector database improves the efficiency of knowledge retrieval and question answering, thereby reducing calculation costs and time, and improving calculation accuracy. In combination with a digital sand table, visual interaction of an optimization design scheme is realized.

Description

A decision-making method for optimizing the non-uniform indoor environment of tall, spacious buildings based on a multimodal knowledge graph-enhanced base model. Technical Field

[0001] This invention belongs to the field of green building performance optimization design technology, and in particular relates to a decision-making method for optimizing the non-uniform indoor environment of tall, spacious buildings based on a multimodal knowledge graph-enhanced base model. Background Technology

[0002] In the AEC (Autonomous Energy Conservation) field, building energy efficiency requirements have gradually increased from 30% to 91.25%. The "Four Good Construction" series of standards focuses on seven dimensions across multiple scales: livability, intelligence, humanism, safety, comfort, greenness, and service. Faced with these higher demands for building energy efficiency, traditional optimization design decision-making methods rely on performance simulation based on green performance data. However, these methods have limitations in computational efficiency and cost, making it difficult to simultaneously process multimodal building green performance data. They also fail to effectively form multi-platform collaboration between design and simulation, resulting in energy-saving benefits that cannot meet new development needs and effectively respond to the multi-dimensional requirements of buildings.

[0003] The indoor environment of tall, open-plan buildings is non-uniform, influenced by multiple coupled factors, resulting in complex airflow processes and variable environmental parameters. This can easily lead to problems such as heat accumulation and poor air circulation. Different functional areas within such buildings exhibit significant differences in ventilation, heat load, and occupant density, with some areas prone to localized stagnation, impacting user comfort. Achieving effective temperature, humidity, and ventilation control, as well as user comfort adjustment, is a core issue in optimizing the green performance of such buildings. Traditional design decision-making methods struggle to adapt to changing times and locations, failing to effectively respond to higher comfort and energy-saving requirements.

[0004] Current design decisions for optimizing non-uniform indoor environments in tall, spacious buildings are primarily based on physical models and machine learning models, which have the following drawbacks:

[0005] 1. It is difficult to process and integrate multi-source, multi-modal heterogeneous data from non-uniform indoor environments. Traditional methods rely on static data and models, cannot provide feedback on real-time data, lack the ability to respond to emergencies, have slow design response and delayed decision-making, and lack flexibility and adaptability;

[0006] 2. It is difficult to accurately simulate and analyze green performance data under the influence of multiple coupled factors in the non-uniform indoor environment of tall, spacious buildings. Traditional methods lack knowledge reasoning capabilities, cannot effectively leverage the advantages of cloud computing power, have poor learning ability from historical cases and similar solutions, and are difficult to effectively propose personalized design decisions for specific scenarios, resulting in poor flexibility and scalability.

[0007] In recent years, multimodal large language models (MM-LLMs) have demonstrated great potential in multimodal data fusion processing. Based on inputs from modalities such as numerical data, text, images, video, and audio, large models can understand user needs and generate decision suggestions and technical reports. However, the illusion problem of large models limits their accuracy and reliability in engineering scenario design applications. This problem can be effectively improved by utilizing knowledge graph augmentation. Relying on knowledge graphs tailored to specific engineering scenarios, and through retrieval-enhanced generative models, design questions are matched with triples in the knowledge graph, empowering the intelligent question-answering process of large models. Summary of the Invention

[0008] The purpose of this invention is to solve the problems in the prior art and propose a decision-making method for optimizing the non-uniform indoor environment of tall buildings based on a multimodal knowledge graph-enhanced base model.

[0009] This invention is achieved through the following technical solution: This invention proposes a method for optimizing the design of non-uniform indoor environments in tall, spacious buildings based on a multimodal knowledge graph-enhanced base model. The method includes the following steps:

[0010] S1. Acquire knowledge of multimodal data related to non-uniform indoor environments;

[0011] Step S1 includes the following steps:

[0012] S11. Collect green performance data of non-uniform indoor environments;

[0013] S12. Generate an embedded representation that integrates multimodal non-uniform indoor environmental green performance information;

[0014] S13. Extract green performance data information of non-uniform indoor environment;

[0015] S2. Integrate multimodal data knowledge of non-uniform indoor environments to construct a multimodal knowledge graph and vector database of non-uniform indoor environments;

[0016] Step S2 includes the following steps:

[0017] S21. Convert the extracted information into RDF triples;

[0018] S22. Construct a multimodal knowledge graph of non-uniform indoor environment based on RDF triples;

[0019] S23. Determine the elements of typical non-uniform indoor environment engineering scenarios;

[0020] S24. Construct and update the corresponding multimodal knowledge graph and vector database;

[0021] S3. Enhance knowledge retrieval for optimization design decision-making problems and construct a green performance optimization design model that fits typical engineering scenarios of non-uniform indoor environments;

[0022] Step S3 includes the following steps:

[0023] S31. Perform modular clustering of knowledge in the knowledge graph and obtain structured subgraphs;

[0024] S32. Understanding design problems and improving knowledge retrieval efficiency based on structured subgraphs;

[0025] S33. Construct a green performance optimization design model that fits typical engineering scenarios of non-uniform indoor environments;

[0026] S4 enables edge-cloud collaboration to flexibly handle optimization design issues and achieves visualized interaction of optimization design results through a digital sandbox.

[0027] Step S4 includes the following steps:

[0028] S41. Construct an edge-cloud collaborative architecture that fits typical engineering scenarios of various non-uniform indoor environments;

[0029] S42. Construct a digital twin model of a tall, spacious building and display the optimized design results through a digital sand table to achieve visual interaction.

[0030] Further, in step S13, 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 the unstructured, non-uniform indoor environment green performance dataset. Relation extraction refers to using OpenIE and BERT relation classification models or dependency parsing methods to identify the relationships between the aforementioned non-uniform indoor environment green performance entities. Event extraction refers to detecting the spatiotemporal dimension parameters of multiple non-uniform indoor environment green performance entities using the ACE event extraction framework and identifying non-uniform indoor environment green performance event trigger words and related information based on a deep learning model.

[0031] Furthermore, in step S21, the non-uniform indoor environment green performance data is transformed into "entity-relationship-entity" RDF triples using Python's RDF toolkit RDFLib, and the triples are stored in a graph database for subsequent processing and query operations.

[0032] In step S22, the extracted entities are aligned with existing entities in the non-uniform indoor environment to construct a relationship graph between entities, and a graph matching algorithm is used to link the entities. For records in the non-uniform indoor environment 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. Relationship similarity is calculated using an embedding model, and the relationships between the same entities from different data sources are integrated using a graph neural network. The integrated data is then used for rule-based reasoning, based on priority rules set according to the reliability and timestamp of the data sources, to resolve conflicting or contradictory knowledge obtained from different data sources and derive new knowledge. Based on machine learning reasoning, the implicit relationships of green performance data in the non-uniform indoor environment are mined to enhance knowledge, improve the accuracy of information extraction and the reasoning ability of the knowledge graph.

[0033] Further, in step S24, the extracted non-uniform indoor environment multimodal data is preprocessed; the structured data of real-time monitoring values ​​of temperature, humidity, air velocity, light intensity, and energy consumption are normalized and dimensionality reduced; the text data of project information, user requirements, design specifications and standards, and policy documents are generated into embedding vectors based on a natural language processing model; the image data of building photos, renderings, site plans, floor plans, sections, elevations, detailed drawings, and pseudo-color images of optimization simulation results are extracted using ResNet or CNN models; and the user behavior videos and building 3D model videos are extracted from keyframes using C3D networks and LSTM models.

[0034] In step S24, based on the generated embedding vectors, the feature dimensions of green performance data are standardized; an HNSW vector index structure is constructed to improve the speed of vector similarity search and provide support for matching semi-structured data in multimodal knowledge graphs, thereby accelerating the intelligent question-and-answer process for user-designed questions.

[0035] In step S24, for the green performance evaluation indicators of typical engineering scenarios, the influencing factors are further determined, including equipment distribution, wall insulation performance, natural lighting, personnel flow and dynamic load; and the typical model or algorithm elements in the calculation or simulation are identified, including CFD model, solar radiation analysis model, dynamic energy consumption analysis model, PMV model, PPD model and energy consumption optimization algorithm, regional zoning control algorithm;

[0036] In step S24, a green performance profile of the non-uniform indoor environment of a specific tall space engineering scenario is constructed, and the multimodal green performance database of the specific tall space engineering scenario is completed and updated.

[0037] Furthermore, in step S31, the knowledge graph is segmented 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.

[0038] Furthermore, in step S33, based on the multimodal knowledge graph that fits the typical engineering scenario of green performance optimization design for non-uniform indoor environments, an enhanced base model empowered by knowledge representation of green performance in non-uniform indoor environments is trained on the cloud platform based on the base model. This leverages the advantages of cloud computing power to handle the coupling relationships between multimodal data involved in the optimization design task of non-uniform indoor environments, and empowers decision-making modes based on the designer's subjective experience.

[0039] Furthermore, in step S41, based on the green performance optimization design scenario text and photo data input by the designer, and taking into account the parameters of temperature, humidity, air quality, and air velocity, as well as the heat source distribution, personnel distribution, and usage scenario constraints, the design problem of non-uniform indoor environment optimization is understood, and typical problems of non-uniform environment optimization design for tall buildings are addressed.

[0040] In step S41, based on the edge-cloud collaborative distributed working architecture, for the complex green performance design decision problem of non-uniform indoor environment, the data processing capability of the enhanced base model empowered by cloud knowledge representation is used to obtain a comprehensive understanding of the multimodal performance data fusion characteristics; and in collaboration with various building design and building simulation software or platforms, green performance optimization design decisions for non-uniform indoor environment of tall space buildings are formulated, overcoming the limitations of specific algorithms or software tools in terms of computational efficiency or computational cost;

[0041] In step S41, when the designer inputs optimization design task instructions into the large model, the cloud-side model understands the user's needs and issues simulation instructions to the edge model to guide the edge model in simulation. The edge model combines the cloud-side instructions with the engineering scenario model or non-uniform indoor environment parameters to perform green performance simulation optimization, output simulation results, and provide decision-making basis for the cloud-side model.

[0042] Furthermore, in step S42, a digital twin model of the tall, spacious building is constructed; the changes in parameters and energy consumption caused by the modification of the optimization scheme are analyzed in real time through cloud computing, and VR and AR technologies are combined to realize the visual comparison of the optimization design scheme, and support real-time collaborative adjustment of the optimization design scheme by multiple parties.

[0043] 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 above-mentioned method for optimizing the design decision of non-uniform indoor environment of tall space buildings based on a multimodal knowledge graph-enhanced base model.

[0044] 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 above-described method for optimizing the design of non-uniform indoor environments in tall, spatial buildings based on a multimodal knowledge graph-enhanced pedestal model.

[0045] The beneficial effects of this invention are:

[0046] To address the complexity of non-uniform environments in tall, spacious buildings, this paper emphasizes the multimodal characteristics of data (such as numerical values, text, images, and videos). It proposes a decision-making method for optimizing the non-uniform indoor environment of tall, spacious buildings based on a multimodal knowledge graph-enhanced foundation model. This method involves knowledge representation of multimodal data related to the non-uniform indoor environment of tall, spacious buildings, combined with specific application engineering scenarios, to construct a multimodal knowledge graph and vector database for the green performance of non-uniform indoor environments. This allows for effective representation and storage of multimodal data. During the design decision-making process, a green performance optimization design model tailored to typical engineering scenarios of non-uniform indoor environments in tall, spacious buildings is constructed. This model helps to understand the optimization design problem of non-uniform indoor environments and proposes an optimization design decision-making method. The output includes numerical values ​​of building morphology parameters, spatial parameters, HVAC system parameters, and green performance image data of typical cross-sections of tall, spacious buildings. Building upon this foundation, a method for constructing an enhanced foundation model empowered by knowledge representation is proposed. Based on an edge-cloud collaborative distributed working architecture, it achieves optimal allocation of computing resources, and improves knowledge retrieval and question answering efficiency, reduces computational costs and time, and increases computational accuracy through collaboration with multimodal knowledge graphs and vector databases. Combined with a digital sandbox, it enables visualized interaction of optimized design schemes. Attached Figure Description

[0047] 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.

[0048] Figure 1 is a flowchart of the optimization design decision method for non-uniform indoor environment of tall spatial buildings based on a multimodal knowledge graph-enhanced base model, as described in this invention.

[0049] Figure 2 is a block diagram of a method for optimizing the design of non-uniform indoor environments in tall, spacious buildings based on a multimodal knowledge graph-enhanced base model, as described in this invention. Detailed Implementation

[0050] 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.

[0051] Referring to Figures 1 and 2, this invention proposes a method for optimizing the design of non-uniform indoor environments in tall, spacious buildings based on a multimodal knowledge graph-enhanced base model. The method includes the following steps:

[0052] S1. Acquire multimodal data knowledge of non-uniform indoor environment; comprehensively apply LoRa-based MEMS thermal environment monitoring sensor arrays fixed on building facades or installed on intelligent cleaning or service robots, thermal imaging cameras mounted on drones, and multi-wavelength high-density point cloud LiDAR and other indoor environment physical parameter acquisition devices, as well as high-resolution cameras, multispectral imaging quantum dot infrared imagers, ultra-wideband (UWB) multiple-input multiple-output (MIMO) radar, flexible high-speed organic photodetectors, and other user data acquisition devices to collect and model multimodal data of non-uniform indoor environment in tall and large-space buildings.

[0053] Step S1 includes the following steps:

[0054] S11. Collect green performance data of non-uniform indoor environments; In step S11, based on the dynamic acquisition of four types of data information—numerical, text, image, and video—of non-uniform indoor environments from integrated platforms such as BIM and IoT, a green performance dataset of non-uniform indoor environments is constructed, and the data modality types and corresponding data format requirements are determined during the data parsing process. BIM platform data is acquired in IFC format, and IoT platform data is acquired in JSON format. Numerical data is input in formats such as CSV and JSON; text data is input in formats such as TXT, CSV, and JSON; image data is input in formats such as JPEG and PNG; and video data is input in formats such as MP4 and AVI.

[0055] S12. Generate an embedded representation that integrates multimodal non-uniform indoor environmental green performance information; use cross-modal alignment algorithms such as CLIP and MMV and multimodal Transformer to process building green performance data of different modalities, and convert these data into high-dimensional vectors that retain their multimodal data features, embed them into a shared vector space to achieve information fusion and alignment, retain and strengthen the correlation between different modal building green performance data, 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 when processing multimodal tasks.

[0056] S13. Extract green performance data information of non-uniform indoor environment; In step S13, 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 the unstructured non-uniform indoor environment green performance dataset; Relation extraction refers to using OpenIE and BERT relation classification models or dependency parsing methods to identify the relationships between the aforementioned non-uniform indoor environment green performance entities; Event extraction refers to detecting the spatiotemporal dimension parameters of multiple non-uniform indoor environment green performance entities through the ACE event extraction framework, and identifying non-uniform indoor environment green performance event trigger words and related information based on deep learning models such as BiLSTM-CRF and BERT.

[0057] S2. Integrate multimodal data knowledge of non-uniform indoor environments to construct a multimodal knowledge graph and vector database for non-uniform indoor environments; apply multidimensional knowledge graphs to develop knowledge representations of data such as temperature, humidity, noise level, air quality, air velocity, pedestrian distribution, and passenger behavior in non-uniform indoor environments of tall buildings; combine engineering scenario types such as industrial plants, stadiums, airport terminals, and large convention centers to construct a multimodal knowledge graph and vector database for the green performance of non-uniform indoor environments of tall buildings.

[0058] Step S2 includes the following steps:

[0059] S21. Convert the extracted information into RDF triples; In step S21, the non-uniform indoor environment green performance data is converted into "entity-relationship-entity" RDF triples using the RDF toolkit RDFLib of Python, and the triples can be stored in a graph database for subsequent processing and querying operations.

[0060] S22. Construct a multimodal knowledge graph of non-uniform indoor environment based on RDF triples. In step S22, the extracted entities are aligned with existing entities in the non-uniform indoor environment to construct a relationship graph between entities. Graph matching algorithms such as PageRank and Random Walk are used to link entities. For records in the non-uniform indoor environment 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 relationships between the same entities from different data sources are integrated using graph neural networks such as GCN and RGCNs. The integrated data is then used for rule-based reasoning. Based on the reliability of the data source, timestamps, and other pre-defined priority rules, conflicting or contradictory knowledge obtained from different data sources is resolved, and new knowledge is derived. Reasoning based on machine learning such as GNNs is used to mine implicit relationships in the green performance data of the non-uniform indoor environment, thereby enhancing knowledge, improving the accuracy of information extraction, and enhancing the reasoning ability of the knowledge graph.

[0061] S23. Determine the typical non-uniform indoor environment engineering scenario elements; for tall, spacious buildings with non-uniform indoor environment characteristics, such as industrial plants, stadiums, airport terminals, and large convention centers, identify their engineering scenario types and extract corresponding green performance evaluation indicators, including power intensity, equipment energy consumption ratio, building energy consumption, indoor air quality, thermal comfort, and lighting energy efficiency.

[0062] S24. Construct and update the corresponding multimodal knowledge graph and vector database. In step S24, the extracted non-uniform indoor environment multimodal data is preprocessed. Real-time monitoring numerical structured data such as temperature, humidity, air velocity, light intensity, and energy consumption are normalized and dimensionality reduced. Textual data such as project information, user requirements, design specifications and standards, and policy documents are used to generate embedding vectors based on natural language processing models such as BERT and TF-IDF. Image data such as building photos, renderings, site plans, floor plans, sections, elevations, detailed drawings, and pseudo-color images of optimization simulation results are used to extract feature vectors using ResNet and CNN models. Video data such as user behavior videos and 3D building models are used to extract features from keyframes using C3D networks and LSTM models. Based on the generated embedding vectors, the feature dimensions of green performance data are standardized. An HNSW vector index structure is constructed to improve the speed of vector similarity search and provide support for matching semi-structured data in the multimodal knowledge graph, accelerating the intelligent question-and-answer process for user design questions. For the green performance evaluation indicators of the aforementioned typical engineering scenarios, further identify their influencing factors, including equipment distribution, wall insulation performance, natural lighting, personnel flow, and dynamic load; and clarify the typical models or algorithms used in calculation or simulation, including CFD models, solar radiation analysis models, dynamic energy consumption analysis models, PMV models, PPD models, energy consumption optimization algorithms, and regional zoning control algorithms. Construct a green performance profile of the non-uniform indoor environment of specific tall and large-space engineering scenarios, and complete and update the multimodal green performance database for specific tall and large-space engineering scenarios.

[0063] S3. Enhance knowledge retrieval for optimization design decision-making problems and construct a green performance optimization design model that fits typical engineering scenarios of non-uniform indoor environments; based on the multimodal knowledge graph and vector database of non-uniform indoor environments, construct a green performance optimization design model that coordinates the indoor environment physical parameter acquisition equipment, the user data acquisition equipment, HVAC equipment, intelligent operation and maintenance robots, edge computing equipment, and cloud computing platforms for typical engineering scenarios of non-uniform indoor environments.

[0064] Step S3 includes the following steps:

[0065] S31. Modular clustering of knowledge in the knowledge graph to obtain structured subgraphs; in step S31, a community detection algorithm is used to identify interconnected node groups, dividing the knowledge graph into different communities. A graph embedding algorithm is used to identify and add missing relationships or entities, enhancing the completeness of the knowledge graph and assisting knowledge reasoning.

[0066] S32. Understanding the Design Problem and Improving Knowledge Retrieval Efficiency Based on Structured Subgraphs: In step S32, based on the green performance optimization design engineering scenario data such as the design problem text, real-world photos, and plan, elevation, and section engineering drawings input by the designer, the potential green performance hazards of the specific optimization design scenario are understood. Considering indoor physical environment parameters such as temperature, humidity, illuminance, air quality, air pollutant concentration, and ventilation volume, and taking into account limiting factors such as equipment distribution, wall insulation performance, natural lighting, personnel flow, and dynamic load, the non-uniform indoor environment optimization design problem is understood. Hierarchical clustering methods such as hierarchical clustering and k-means clustering are used to extract summaries from typical engineering scenarios of multi-dimensional non-uniform indoor environments, and the communities are analyzed hierarchically. Algorithms such as Node2Vec, DeepWalk, and GraphSAGE are used to convert the structured subgraphs and their summaries into high-dimensional vector representations. Algorithms such as FAISS and HNSW are used to assist in high-dimensional vector retrieval, improving the efficiency of machine learning tasks such as clustering and retrieval.

[0067] S33. Construct a green performance optimization design model that fits typical engineering scenarios of non-uniform indoor environments. In step S33, based on the multimodal knowledge graph that fits typical engineering scenarios of green performance optimization design for non-uniform indoor environments, and combined with multimodal neural network models such as Transformer, GNN, and RNN, the coupling relationship between multimodal data involved in the optimization design task of non-uniform indoor environments is processed, empowering the traditional decision-making mode based on the designer's subjective experience. In step S33, the main output data includes numerical and image data to provide design alternatives at the levels of building form, spatial layout, and energy system design, responding to the climate, environmental, and cultural requirements of the design task, and assisting in the formulation of optimization design decisions. In step S33, the model collaborates with various building design and building simulation software or platforms to process multimodal building green performance data, overcoming the limitations of specific algorithms or software tools in terms of computational efficiency or computational cost. In step S33, the output numerical data includes building form parameters such as building shape coefficient and window-to-wall ratio, building space parameters such as bay width, depth and floor height, and HVAC system parameters such as equipment operating time and power consumption; the output image data includes green performance predictions for typical cross-sections of tall and large-space buildings, such as temperature gradient prediction maps, flow field prediction maps, and glare simulation maps.

[0068] The green performance optimization design model for typical engineering scenarios of non-uniform indoor environments includes:

[0069] The monitoring module includes a physical parameter acquisition submodule and a user data acquisition submodule. The physical parameter acquisition submodule includes: a LoRa-based MEMS thermal environment monitoring sensor array fixed to the building facade or installed on intelligent cleaning or service robots to collect data such as temperature, humidity, air quality, and air velocity in the non-uniform indoor environment of tall buildings; a pipe inspection robot that uses television imaging to detect cracks, blockages, and corrosion in HVAC equipment pipes; a high-performance wheeled robot that detects abnormalities such as fires and smoke and monitors indoor safety; and a drone that monitors and analyzes the thermal distribution of the building facade of tall buildings. The user data acquisition submodule includes: a high-resolution camera that uses computer vision algorithms to identify human movement; a multispectral imaging quantum dot infrared imager that monitors the density of people in the space in real time; an ultra-wideband (UWB) multiple-input multiple-output (MIMO) radar that acquires high-quality images of respiratory signals in multi-person scenarios such as entrance halls, security checkpoints, waiting areas, and commercial areas; and flexible high-speed organic photodetectors that collect physiological signals such as respiration, heart rate, and skin conductance of users in office areas.

[0070] The retrieval module leverages cross-modal alignment technology and the data processing capabilities of the cloud platform to gain a comprehensive understanding of the fusion characteristics of multimodal performance data and design problems. For user-inputted design problems, based on "knowledge retrieval" and "retrieval enhancement generation" technologies, the design problems are vectorized and compared with the multimodal knowledge graph and vector database for similarity retrieval. The knowledge graph is used to structure semantic problems, determine the relationships between feature vectors and entities, relations, and attributes in the graph database; the vector database stores data feature vectors to find the most relevant matches for green performance multimodal data. Based on semantically similar cases and other knowledge, the module enhances its ability to capture key design problems, generates computational simulation instructions, and clarifies the structured subgraphs of the knowledge graph involved in the optimization design problem.

[0071] The computation module, in collaboration with the retrieval module, integrates the element features of typical engineering scenarios of tall, spacious buildings based on the computational simulation instructions and the structured subgraph of the knowledge graph. It utilizes lightweight, personalized models deployed on the edge to achieve layered lightweight decision-making, dynamically analyzes the green performance evaluation indicators of non-uniform indoor environments, and, based on data collection, integrates multimodal data. It combines thermal simulation software to simulate heat transfer, hot air flow, and light intensity within buildings, as well as CFD simulations of airflow patterns, heat distribution, and temperature gradients. Responding to indoor environments under different seasons and usage scenarios, it analyzes and formulates optimized design decisions for air conditioning design, ventilation system layout, and heat source distribution, outputting the corresponding building form, space, and HVAC system design.

[0072] Interactive Module: Combining the output of the calculation module, a digital twin model of the tall, spacious building is constructed. Incorporating VR and AR technologies, decision support data images such as temperature, humidity, air quality, air velocity, pedestrian flow distribution, and visitor behavior from typical floor plans, elevations, and sections are projected onto the corresponding model surface. Designers can manipulate the digital sandbox by wearing VR devices or moving equipment components within the physical model. The cloud platform calculates and analyzes the changes in parameters and energy consumption caused by these operations in real time, and marks recommended equipment placement locations in alternative optimized design schemes for designers within the digital twin model. This enables visual comparison of optimized design schemes and supports real-time collaborative adjustments to optimized design schemes by multiple parties.

[0073] The digital twin model of the tall, spacious building includes: numerical and image data such as temperature, humidity, air quality, and air velocity of the non-uniform indoor environment of the tall, spacious building output by a MEMS thermal environment monitoring sensor array; image data of pedestrian density output by an infrared imager; numerical data such as building shape coefficient and window-to-wall ratio, and building space parameters such as bay width, depth, and floor height output by LiDAR; image data such as temperature gradient prediction map and flow field prediction map of the tall, spacious building, and green performance prediction for typical cross-sections of the tall, spacious building output by thermal simulation and CFD simulation; and data such as the operating time and power consumption of the HVAC equipment.

[0074] S4. Construct a cloud platform model for non-uniform indoor environment design decision-making, coordinating cloud-side and edge-side models to flexibly handle optimization design problems; based on the non-uniform indoor environment optimization design model, integrate wind field maps of plan, elevation, and section of tall buildings collected by MEMS thermal environment monitoring sensor arrays, thermal maps generated by infrared imagers capturing thermal radiation, 3D models of tall buildings generated based on high-precision LiDAR spatial data, thermal simulation software, and CFD simulation results; construct a visualized digital twin model of the non-uniform indoor environment performance of tall buildings based on VR and AR technologies; and realize the designer's visual interaction with the layout scheme of HVAC equipment in typical functional spaces such as entrance halls, security checkpoints, waiting areas, commercial areas, and office areas through a digital sand table.

[0075] Step S4 includes the following steps:

[0076] S41. Construct an edge-cloud collaborative architecture that fits typical engineering scenarios of various non-uniform indoor environments.

[0077] In step S41, based on the basic information of green performance optimization design scenarios such as text and photos input by the designer, the design considers parameters such as temperature, humidity, air quality, and air velocity, and takes into account limiting factors such as heat source distribution, personnel distribution, and usage scenarios to understand the optimization design problem of non-uniform indoor environment and respond to typical problems of optimization design of non-uniform environment in tall buildings.

[0078] Based on an edge-cloud collaborative distributed architecture, this approach addresses the complex green performance design decision-making problem for non-uniform indoor environments. Leveraging the data processing capabilities of the enhanced base model empowered by cloud-based knowledge representation, it achieves a comprehensive understanding of the fusion characteristics of multimodal performance data. By collaborating with various architectural design and simulation software or platforms, it enables the formulation of optimized green performance design decisions for non-uniform indoor environments in tall, spacious buildings, overcoming the limitations of specific algorithms or software tools in terms of computational efficiency or cost.

[0079] When designers input optimization design task instructions into the large model, the cloud-side model understands the user's needs and issues simulation instructions to the edge model to guide the edge model in simulation. The edge model combines the cloud-side instructions with engineering scenario models or non-uniform indoor environmental parameters to perform green performance simulation optimization, providing the cloud-side model with simulation results and other decision-making basis.

[0080] S42. Construct a digital twin model of a tall, spacious building and display the optimized design results through a digital sand table to achieve visual interaction.

[0081] In step S42, a digital twin model of the tall, spacious building is constructed; the changes in parameters and energy consumption caused by the modification of the optimization scheme are analyzed in real time through cloud computing, and VR and AR technologies are combined to realize the visual comparison of the optimization design scheme and support real-time collaborative adjustment of the optimization design scheme by multiple parties.

[0082] For tall, spacious buildings with non-uniform indoor environments, such as industrial plants, stadiums, airport terminals, and large convention centers, this project integrates data from LoRa-based MEMS thermal environment monitoring sensor arrays, cameras, and infrared imagers. By analyzing the temporal trends of sensor data and peak periods of personnel flow, the project analyzes the temporal and spatial characteristics of engineering scenarios. Based on deep learning models, historical data is analyzed to identify and classify engineering scenario types, extracting green performance evaluation indicators, determining influencing factors, and identifying typical models or algorithms used in calculations or simulations. The resulting multimodal green performance database is then completed and updated and stored in a graph database. Using deep learning algorithms for processing graph-structured data, the interconnected green performance data is traversed to establish a unified representation method for cross-scale multimodal information in the building domain, including text, raster images, 3D models, planar vectors, and building performance. The input discrete data is converted into high-dimensional vectors in a unified multimodal shared vector space. Based on the engineering scenarios, a multimodal knowledge graph is constructed that fits typical engineering scenarios for optimizing green performance in non-uniform indoor environments. Furthermore, by generating embedded vectors, standardizing the feature dimensions of green performance data, constructing an HNSW vector index structure, improving the speed of vector similarity search, and building a vector database of green performance data for non-uniform indoor environments, this provides support for matching semi-structured data in multimodal knowledge graphs.

[0083] The indoor environmental monitoring sensor array includes: a resistive flexible temperature sensor, a capacitive flexible humidity sensor, a tunable diode laser absorption spectroscopy sensor for monitoring the concentration of gases such as CO2, CO, and methane, an optical particle counter for monitoring suspended particulate matter such as PM2.5 and PM10 in the air, and a particle image velocimeter for monitoring airflow velocity.

[0084] The knowledge graph and vector database support multimodal data input, including numerical data (such as indoor environmental data, building form data, green performance data, etc.), text (project information, user requirements, design specifications and standards, policy documents, etc.), images (building photos, renderings, site plans, floor plans, sections, elevations, detailed drawings, pseudo-color images of optimization simulation results, etc.), and videos (user behavior videos, 3D building models, etc.).

[0085] The green performance evaluation indicators include pollutant concentration, the time and area ratio of indoor environmental parameters in the adaptive thermal comfort zone under natural or combined ventilation conditions of main functional rooms, the temperature range of the inner surface of the exterior walls and roof, the average number of natural ventilation air changes in main functional rooms under typical conditions in the transition season, energy consumption of cold and heat sources, energy consumption of transmission and distribution systems, renewable energy utilization rate, and electricity intensity.

[0086] Typical problems in the optimization design of the indoor environment of tall buildings include: uneven distribution of longitudinal heat sources in tall spaces, requiring a balance between energy consumption and comfort, analysis of heat convection, and determination of local heating and fresh air system solutions; large temperature and humidity gradients caused by height differences, requiring coordination of temperature and humidity and development of zoning design schemes for air conditioning systems; and uneven airflow, requiring the identification of air stagnation zones, installation of air circulation fans, and a combination of natural and mechanical ventilation to overcome spatial height differences and airflow stratification.

[0087] Example

[0088] Figure 1 illustrates a non-uniform indoor environment optimization design decision-making method based on a multimodal knowledge graph-enhanced base model, applicable to tall, open-air buildings such as airport terminals. The terminal's indoor environment is complex, with uneven passenger flow distribution, densely populated areas with alternating functions, and significant variations in usage frequency and demand. The indoor temperature and humidity distribution is also uneven, and issues such as heat accumulation and poor air circulation are likely to occur. The design must balance energy conservation and comfort while ensuring smooth passenger flow, achieving intelligent control.

[0089] S1. Acquire knowledge of multimodal data related to non-uniform indoor environments;

[0090] In this embodiment, data collection is based on BIM and IoT platforms. Data collected using the BIM platform includes: geometric spatial data such as building dimensions, location, spatial layout, and window-to-wall ratio; material physical properties such as thermal conductivity, thermal resistance, and acoustic properties of building materials; basic parameters and pipeline layout of the building's HVAC system and other electrical equipment; energy monitoring data such as daylighting and energy consumption; and building operation and maintenance data such as equipment maintenance records and operation schedules. Based on the IoT platform, real-time dynamic data of the terminal building can be obtained, specifically including: indoor physical environment parameters such as temperature and humidity, air quality, noise level, and illuminance on work surfaces; energy consumption monitoring data such as real-time total energy consumption of HVAC system equipment or real-time energy consumption of specific areas and specific equipment; and user-related data such as pedestrian flow distribution and passenger behavior. Simultaneously, it can also record dynamic changes in outdoor physical environment parameters such as meteorological data, wind speed, and wind direction, and supplement project information, user requirements, design specifications and standards, and policy documents.

[0091] The geometric spatial data includes parametric numerical data, descriptive text data, and image data such as architectural photographs, renderings, site plans, floor plans, sections, elevations, and detailed drawings. The material physical properties include numerical data and descriptive text data. The energy monitoring data includes numerical data with time-series information, descriptive text data, pseudo-color images, and video data. The indoor physical environment data includes numerical data with time-series information and pseudo-color images. The user-related data includes questionnaire text data, photographs, and behavioral videos.

[0092] The raw data from the different data sources are preprocessed to extract structured information. Numerical data in tabular or similar formats does not require this processing. For text data, semantic segmentation and part-of-speech tagging are performed based on NLP or LLM models.

[0093] Utilizing a multimodal Transformer deep learning architecture, and employing models such as ViLBERT and UNITER, data from different modalities—numerical, text, image, and video—are embedded into vector spaces using different encoders. A self-attention mechanism is employed to allocate weights across different modalities, enhancing the understanding of data relationships between them, such as the relationship between an airport terminal cross-section and its corresponding annotation text. Representations relevant to both modalities are generated, while preserving the temporal information of dynamic data.

[0094] For the preprocessed data, entity and relation data are extracted. Deep learning-based models such as BERT, RoBERTa, and spaCy are applied to classify and identify entities. These entities contain category information related to the green performance of the terminal building or related concepts, forming the subject or object of RDF triples. Relational classification models such as OpenIE and BERT, or dependency parsing methods, are used to identify the relationships between the aforementioned green performance entities of the terminal building's indoor environment, outputting the predicate of the RDF triples.

[0095] For event data involving multiple entities, attributes, and time-series information, the ACE event extraction framework is used to label entities and relationships, and deep learning models such as BiLSTM-CRF and BERT are used to identify event trigger words, roles, and other information, which are then converted into triples.

[0096] S2. Integrate multimodal data knowledge of non-uniform indoor environments to construct a multimodal knowledge graph and vector database of non-uniform indoor environments;

[0097] The green performance data of the terminal's indoor environment was transformed into RDF triples of "entity-relationship-entity" using the RDF toolkit RDFLib in Python and stored in the Neo4j graph database.

[0098] When new knowledge about the green performance of the terminal's indoor environment is extracted, the above steps are repeated to extract the entity, relationship, and attribute information of the knowledge. The newly extracted entities related to the green performance of the terminal's indoor environment are identified using contextual information, tagged, and embedded into the vector space. Links are then established based on semantic similarity calculation. For textual information, entities can be matched based on string similarity. For records in the graph database pointing to the same entity, clustering algorithms such as K-means and DBSCAN, or machine learning methods such as logistic regression and random forest, are used to filter items with similar names or IDs. Pattern matching or rule reasoning is used to determine whether the relationships between different inputs are the same. Graph embedding methods such as DeepWalk and Node2Vec are used to embed the newly extracted entities related to the green performance of the terminal's indoor environment, along with their relationships and attributes, into the vector space. Duplicate entities are identified and merged through distance calculation. Furthermore, semantic similarity calculation is used to determine whether the relationships between entities are consistent, and graph convolutional neural networks or relational attention networks are used to fuse similar relationships from different sources. In cases where entities or attributes conflict, data is sorted based on factors such as the reliability of the terminal data source, timestamps, and value ranges. The data with the highest priority is selected as the reliable data, and ontology reasoning algorithms are used to verify that the logical constraints between the data satisfy objective laws and architectural common sense. Building upon this foundation, the GNN algorithm is used to further infer new knowledge.

[0099] For terminal building projects, considering their complex indoor environment, uneven distribution of people flow, and changes over time, the evaluation indicators include: temperature, relative humidity, air velocity, air quality (carbon dioxide concentration, PM2.5 concentration) in different indoor areas, at different heights, and in typical cross-sections; illuminance on work surfaces; power consumption of HVAC and lighting systems in different areas; equipment energy consumption; user thermal comfort; and changes in people density.

[0100] The distribution of heating and cooling loads in an airport terminal is complex, influenced by its large dimensions of depth, width, and height. Different areas are affected by different factors, including: 1. Proximity to the curtain wall: Areas near the curtain wall are significantly affected by outdoor environmental factors such as solar radiation and infiltration winds, while areas far from the curtain wall are mainly affected by passenger flow and density, and equipment heat dissipation; 2. Clear height: Due to the upward movement of hot air, there are significant temperature differences between vertically connected areas; 3. Different usage patterns: The main areas of the terminal include the entrance hall, check-in area, security checkpoint, waiting area, baggage claim area, and commercial area. Among these, the entrance hall has high passenger flow; the check-in area is characterized by high-density, short-term passenger gatherings; the security checkpoint and baggage claim areas exhibit both passenger congestion and flow patterns; the waiting area has longer passenger congestion times; and the commercial area has fluctuating passenger density.

[0101] The data is preprocessed by normalizing and reducing the dimensionality of real-time monitored numerical structured data such as temperature, humidity, air velocity, light intensity, and energy consumption for different regions or heights and usage scenarios. Feature vectors are extracted from image data such as pseudo-color images of typical area profile optimization simulation results and user behavior videos. Natural language processing is performed on text data such as common design specifications, standards, and policy documents for the terminal building to generate embedding vectors. Feature vectors are extracted from image data such as photos, renderings, site plans, floor plans, sections, elevations, and detailed drawings of specific terminal buildings using ResNet and CNN models. Features are extracted from keyframes of the 3D building model using C3D networks and LSTM models. Based on this, an HNSW vector index structure is constructed to build a vector database of a multimodal knowledge graph of the terminal building's indoor environment, assisting in green performance optimization design decisions for non-uniform indoor environments of tall buildings, such as uneven longitudinal heat source distribution, large temperature and humidity gradients, uneven air flow, local glare, or insufficient lighting.

[0102] S3. Enhance knowledge retrieval for optimization design decision-making problems and construct a green performance optimization design model that fits typical engineering scenarios of non-uniform indoor environments;

[0103] For the knowledge graph of the terminal's green performance data, community detection algorithms such as Louvain and Label Propagation are used for coarse segmentation, efficiently dividing the community structure in a large-scale knowledge graph. Node2Vec and DeepWalk random walks are used to embed text nodes and graph nodes respectively, enhancing the integrity of the knowledge graph and assisting knowledge reasoning.

[0104] Hierarchical clustering methods such as hierarchical clustering and k-means clustering are used to extract the summary of the terminal building project scene. The entities and relationships of the green performance knowledge of the terminal building are converted into high-dimensional vectors by combining Node2Vec, DeepWalk, and GraphSAGE algorithms. The embedded communities and entities are stored in a vector database using algorithms such as FAISS and HNSW to assist in retrieval.

[0105] Based on the green performance optimization design engineering scenario data input during the question-and-answer process, including design question text, real-life photos of specific terminal scenes, terminal plan, elevation, and section engineering drawings, and typical section simulation diagrams of the terminal, the specific optimization design scenario type (entrance hall, check-in area, security checkpoint, waiting area, baggage claim area, commercial area) is determined. Potential green performance hazards in the specific optimization design scenario are understood, and the distribution of temperature, relative humidity, air velocity, and air quality (carbon dioxide concentration, PM2.5 concentration) in the relevant area's plan and typical sections is searched. Also considered are the power consumption and energy consumption of the area's HVAC and lighting systems, as well as user thermal comfort and changes in pedestrian density. For check-in and security checkpoints, illuminance information on the work surfaces needs to be taken into account.

[0106] This system collaboratively utilizes CFD models, solar radiation analysis models, dynamic energy consumption analysis models, PMV models, PPD models, energy consumption optimization algorithms, and regional zoning control algorithms to address specific green performance simulation analysis problems. It couples outdoor physical environment elements with indoor physical environment data from multiple prediction models to support decision-making regarding the green performance of the terminal building and assist designers in comparing different options. It outputs alternative solutions such as optimized local morphology or spatial layout suggestions for the terminal building, optimized design suggestions for energy systems or technology platforms, and supporting data including building morphology parameters such as the terminal building's shape coefficient or window-to-wall ratio, building spatial parameters such as the bay or depth of specific areas after optimizing the spatial layout, HVAC system parameters such as the operating time and power consumption of existing or multiple optimized schemes' regional equipment, and predicted temperature gradient maps, flow field maps, and glare simulation maps for typical cross-sections. This overcomes the limitations of specific algorithms or software tools in terms of computational efficiency or cost.

[0107] S4. Construct a cloud platform model for non-uniform indoor environment design decision-making, and coordinate cloud-side and edge-side models to flexibly handle optimization design problems.

[0108] Based on the dataset obtained from the execution of design simulation and decision-making tasks using the green performance optimization design model that fits typical terminal engineering scenarios, and leveraging the computing power and data processing capabilities of the cloud platform, an enhanced base model empowered by knowledge representation of green performance of the terminal is constructed based on the Mixtral8x78 large model. This model comprehensively understands the multimodal building information involved in the terminal scenario and better supports the reasoning and analysis capabilities of the green performance optimization design model in dealing with complex problems.

[0109] Building upon this foundation, the collaborative analysis capabilities of the cloud-side enhanced base model and the edge-side green performance optimization design model are strengthened. This reduces computational costs and time while improving computational efficiency and accuracy. When designers input optimization design task instructions into the large model, the cloud-side model understands the user's needs and issues simulation instructions to the edge-side model, guiding it to perform simulations. The edge-side model combines the cloud-side instructions with engineering scenario models or non-uniform indoor environmental parameters to perform green performance simulation optimization, outputting simulation results and providing decision-making support for the cloud-side model.

[0110] For real-time updated green performance data of the terminal's indoor environment, such as IoT data, data is collected in real time via IoT sensors. Based on the required measurement accuracy, analysis intervals are determined, data is analyzed, and corresponding decision-making reference solutions are provided. Designers or terminal operation and maintenance personnel compare or modify the alternative solutions to formulate operational plans for indoor physical environment control equipment. After excluding outlier data, statistical parameters such as the average and standard deviation of the data over different time periods are analyzed to summarize periodic patterns and characteristics. When the deviation exceeds a preset threshold, the knowledge graph is updated.

[0111] In the process of optimizing the green performance design, a green performance profile of the terminal's indoor environment is constructed, and the multimodal green performance database of the terminal is completed and updated.

[0112] 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 above-mentioned method for optimizing the design decision of non-uniform indoor environment of tall space buildings based on a multimodal knowledge graph-enhanced base model.

[0113] 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 above-described method for optimizing the design of non-uniform indoor environments in tall, spatial buildings based on a multimodal knowledge graph-enhanced pedestal model.

[0114] 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 (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). 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.

[0115] 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)).

[0116] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0117] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The 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 devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0118] The above provides a detailed description of the optimization design decision-making method for non-uniform indoor environments of tall, spacious buildings based on a multimodal knowledge graph-enhanced base 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 the design of non-uniform indoor environments in tall, spacious buildings based on a multimodal knowledge graph-enhanced base model, characterized in that, The method includes the following steps: S1. Acquire knowledge of multimodal data related to non-uniform indoor environments; Step S1 includes the following steps: S11. Collect green performance data of non-uniform indoor environments; S12. Generate an embedded representation that integrates multimodal non-uniform indoor environmental green performance information; S13. Extract green performance data information of non-uniform indoor environment; S2. Integrate multimodal data knowledge of non-uniform indoor environments to construct a multimodal knowledge graph and vector database of non-uniform indoor environments; Step S2 includes the following steps: S21. Convert the extracted information into RDF triples; S22. Construct a multimodal knowledge graph of non-uniform indoor environment based on RDF triples; S23. Determine the elements of typical non-uniform indoor environment engineering scenarios; S24. Construct and update the corresponding multimodal knowledge graph and vector database; S3. Enhance knowledge retrieval for optimization design decision-making problems and construct a green performance optimization design model that fits typical engineering scenarios of non-uniform indoor environments; Step S3 includes the following steps: S31. Perform modular clustering of knowledge in the knowledge graph and obtain structured subgraphs; S32. Understanding design problems and improving knowledge retrieval efficiency based on structured subgraphs; S33. Construct a green performance optimization design model that fits typical engineering scenarios of non-uniform indoor environments; S4 enables edge-cloud collaboration to flexibly handle optimization design issues and achieves visualized interaction of optimization design results through a digital sandbox. Step S4 includes the following steps: S41. Construct an edge-cloud collaborative architecture that fits typical engineering scenarios of various non-uniform indoor environments; S42. Construct a digital twin model of a tall, spacious building and display the optimized design results through a digital sand table to achieve visual interaction.

2. The method according to claim 1, characterized in that, In step S13, 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 the unstructured, non-uniform indoor environment green performance dataset. Relation extraction refers to using OpenIE and BERT relation classification models or dependency parsing methods to identify the relationships between the aforementioned non-uniform indoor environment green performance entities. Event extraction refers to detecting the spatiotemporal dimension parameters of multiple non-uniform indoor environment green performance entities using the ACE event extraction framework and identifying event trigger words and related information based on a deep learning model.

3. The method according to claim 1, characterized in that, In step S21, the non-uniform indoor environment green performance data is transformed into "entity-relationship-entity" RDF triples using Python's RDF toolkit RDFLib, and the triples are stored in a graph database for subsequent processing and query operations. In step S22, the extracted entities are aligned with existing entities in the non-uniform indoor environment to construct a relationship graph between entities, and a graph matching algorithm is used to link the entities; for records in the non-uniform indoor environment knowledge base that point to the same entity, a clustering algorithm or a DeepWalk algorithm is used to calculate and merge the distance between entities and relationships embedded in the vector space; 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 integrated data, through rule-based reasoning, resolves conflicting or contradictory knowledge obtained from different data sources and derives new knowledge based on priority rules set according to the reliability and timestamps of the data sources. Based on machine learning reasoning, we can mine the implicit relationships in the green performance data of non-uniform indoor environments, perform knowledge enhancement, and improve the accuracy of information extraction and the reasoning ability of knowledge graphs.

4. The method according to claim 1, characterized in that, In step S24, the extracted non-uniform indoor environment multimodal data are preprocessed; the real-time monitoring numerical structured data of temperature, humidity, air velocity, light intensity and energy consumption are normalized and dimensionality reduced. For textual data such as project information, user requirements, design specifications and standards, and policy documents, embedding vectors are generated based on natural language processing models; For image data such as architectural photos, renderings, site plans, floor plans, sections, elevations, detailed drawings, and pseudo-color images of optimized simulation results, feature vectors are extracted using ResNet or CNN models. For user behavior videos and architectural 3D model videos, features are extracted from keyframes using C3D networks and LSTM models. In step S24, the feature dimensions of the green performance data are standardized based on the generated embedding vectors; Construct an HNSW vector index structure to improve the speed of vector similarity search and provide support for matching semi-structured data in multimodal knowledge graphs, thereby accelerating the intelligent question-answering process for user-designed questions; In step S24, for the green performance evaluation indicators of typical engineering scenarios, the influencing factors are further determined, including equipment distribution, wall insulation performance, natural lighting, personnel flow and dynamic load; and the typical model or algorithm elements in the calculation or simulation are identified, including CFD model, solar radiation analysis model, dynamic energy consumption analysis model, PMV model, PPD model and energy consumption optimization algorithm, regional zoning control algorithm; In step S24, a green performance profile of the non-uniform indoor environment of a specific tall space engineering scenario is constructed, and the multimodal green performance database of the specific tall space engineering scenario is completed and updated.

5. The method according to claim 1, characterized in that, 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.

6. The method according to claim 1, characterized in that, In step S33, based on the multimodal knowledge graph that fits the typical engineering scenario of green performance optimization design for non-uniform indoor environments, an enhanced base model empowered by knowledge representation of green performance in non-uniform indoor environments is trained on the cloud platform based on the base model. This leverages the advantages of cloud computing power to handle the coupling relationship between multimodal data involved in the optimization design task of non-uniform indoor environments, and empowers decision-making modes based on the designer's subjective experience.

7. The method according to claim 1, characterized in that, In step S41, based on the green performance optimization design scenario text and photo data input by the designer, and taking into account factors such as temperature, humidity, air quality, and air velocity, as well as heat source distribution, personnel distribution, and usage scenario constraints, the design problem of non-uniform indoor environment optimization is understood, and typical problems of non-uniform environment optimization design for tall buildings are addressed. In step S41, based on the edge-cloud collaborative distributed working architecture, for the complex green performance design decision problem of non-uniform indoor environment, the data processing capability of the enhanced base model empowered by cloud knowledge representation is used to obtain a comprehensive understanding of the multimodal performance data fusion characteristics. By collaborating with various architectural design and simulation software or platforms, we can formulate green performance optimization design decisions for non-uniform indoor environments of tall and spacious buildings, thus overcoming the limitations of specific algorithms or software tools in terms of computational efficiency or cost. In step S41, when the designer inputs optimization design task instructions into the large model, the cloud-side model understands the user's needs and issues simulation instructions to the edge model to guide the edge model in simulation. The edge model combines the cloud-side instructions with the engineering scenario model or non-uniform indoor environment parameters to perform green performance simulation optimization, output simulation results, and provide decision-making basis for the cloud-side model.

8. The method according to claim 1, characterized in that, In step S42, a digital twin model of the tall, spacious building is constructed; the changes in parameters and energy consumption caused by the modification of the optimization scheme are analyzed in real time through cloud computing, and VR and AR technologies are combined to realize the visual comparison of the optimization design scheme and support real-time collaborative adjustment of the optimization design scheme by multiple parties.

9. 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 1-8.

10. 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 1-8.

Images