BIM rule specification extraction method based on large model

Abstract

The invention discloses a BIM rule specification extraction method based on a large model, and the method comprises the following steps: S1, BIM standard document content extraction: extracting all contents in a document from a BIM standard document, and carrying out the corresponding processing, thereby converting the contents into a form which can be used for the subsequent semantic understanding of the large model and the BIM rule specification extraction; s2, semantic disassembly of BIM standard document content: based on the content extracted in the step S1, extracting entities and relationships thereof in the content; s3, BIM rule specification fusion and processing: based on the extracted entities and the relationships thereof, performing knowledge fusion and knowledge processing to form BIM rule specifications; and S4, BIM rule specification regularization and semantization: through BIM rule specification regularization, realizing system integration of BIM rule specifications from the BIM rule specification knowledge graph to a BIM platform. According to the method, the BIM rule specification is efficiently, rapidly and accurately extracted from the BIM standard, so that the time cost and the error possibility of manual reading and rule document arrangement are greatly reduced, the heavy workload of manual reading understanding and BIM standard execution is solved, the BIM efficiency in an actual engineering project and the BIM standard system implementation efficiency are improved, and the BIM standard system implementation efficiency is improved. And the accuracy of standardizing the BIM rule to an actual engineering project is improved.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to a BIM rule specification extraction method based on a large model. Background Art

[0002] BIM (Building Information Modeling) is the abbreviation of Building Information Modeling. It is a technology widely used in architecture, engineering, construction and facility management to generate and manage digital representations of the physical and functional characteristics of building projects. BIM technology can not only provide a multi-dimensional information model that enables all parties in the project to share and update information throughout the life cycle of the building, but also predict performance, optimize design and reduce costs by simulating the design, construction and operation of the building. As an integrated information processing technology for construction projects, BIM technology supports the design, construction and management of buildings through digital information models, and has become an important tool for improving the efficiency of construction projects, reducing costs and promoting multi-party collaboration.

[0003] The BIM standard system covers BIM rules and specifications for each stage from design to construction to operation and maintenance. BIM rules and specifications are the key to ensuring that BIM is carried out in accordance with unified standards throughout the entire life cycle of the project, ensuring that all participants from design to construction to operation and maintenance can follow unified standards in BIM, and ensuring the accuracy and operability of information.

[0004] The BIM-related standard system is complex and has a wide range of content. It covers many different industries such as highways, bridges, culverts, water transport, construction, rail transit, municipal administration, airports, etc. horizontally, and covers multiple types of standard specifications at various stages of engineering projects such as design, construction, and operation and maintenance vertically, and these standards are constantly being updated and improved. Although the existence of the BIM standard system has greatly promoted the standardization and standardized application of BIM technology, in actual operation, extracting BIM rules and specifications from standard documents and applying them to BIM software is still a complex and time-consuming process. At present, this process mainly relies on manual reading, understanding, and entering BIM rules and specifications in BIM standards. It is done manually by professionals, which is not only inefficient, but also prone to misunderstandings and errors due to human factors. The workload of this method of manually entering and maintaining BIM rules and specifications is extremely huge, which brings challenges to accurate understanding and rapid application.

[0005] With the continuous development and application of BIM technology, higher requirements are put forward for the automated extraction and application of BIM rules and specifications. It is particularly urgent to develop an automated method for extracting BIM rules and specifications.

[0006] The information disclosed in this background technology section is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0007] The purpose of the present invention is to provide a BIM rule specification extraction method based on a large model to solve the problems existing in the prior art.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions: The present invention provides a BIM rule specification extraction method based on a large model, comprising the following steps: S1. Extraction of BIM standard document content: Extract all the contents of the BIM standard document and process them accordingly, so as to transform them into a form that can be used for subsequent semantic understanding of the large model and extraction of BIM rules and specifications; S2. Semantic decomposition of BIM standard document content: Based on the content extracted in step S1, extract the entities and their relationships therein for use in subsequent steps to form BIM rules and specifications; S3. Integration and processing of BIM rules and specifications: Based on the extracted entities and their relationships, BIM rules and specifications are formed through knowledge integration and knowledge processing; S4. Regularization and semanticization of BIM rules and specifications: Through the regularization of BIM rules and specifications, the system integration of BIM rules and specifications from the BIM rules and specifications knowledge graph to the BIM platform is realized, and the user-oriented natural language conversion of BIM rules and specifications is formed.

[0009] Further, the specific implementation steps of S1 are as follows: S11. Convert BIM standard document format to PDF format; S12. Classification, recognition and extraction of document content in PDF format.

[0010] Further, the specific implementation steps of S12 are as follows: S121. Extraction of text paragraphs: First, by parsing the internal structure of the PDF file, the text flow and layout attributes are identified, including font size, line spacing, and text position information, focusing on the spatial distribution and alignment between text blocks; then, the layout information is used to distinguish independent text blocks and organize them into paragraphs according to visual continuity; S122, extracting table content: first, by analyzing the page elements of the PDF document, the basic lines and text blocks constituting the table are identified, and by analyzing the geometric positions and relative relationships of these elements, the table boundaries, row and column structures, and the positions of cells in the page are determined; then, based on the determined cell positions, the text content is extracted, and the alignment, font size, and style information of the text are identified to accurately reconstruct the data in the cell; S123. Image extraction: PDFMiner is used to extract images from documents.

[0011] Furthermore, the specific implementation steps of S2 are as follows: S21. Prompt word library and its management: to ensure that stable and easy-to-process return results can be obtained when calling large models, and to achieve more advanced processing and application; S22, semantic decomposition: By extracting entities and their relationships from the data source, candidate knowledge units are obtained, and then the transformation from entities and their relationships to rules and regulations is further realized in the subsequent process.

[0012] Further, the specific implementation steps of S21 are as follows: S211. Specify the format of the result returned by the large model; S212, point out clear semantic disassembly tasks; S213, thinking chain prompts; S214, the prompt words are separated and defined.

[0013] Further, the specific implementation steps of S22 are as follows: S221. Semantic decomposition of text: First, perform text preprocessing, clean the text, and remove noise such as format errors or irrelevant information such as headers and footers; then, develop a text classification algorithm based on the BERT large model to achieve text decomposability analysis and evaluation; then, based on the results of the preliminary classification, develop an entity relationship extraction model based on the Transformer architecture language large model to automatically identify and extract entities and their relationships from the classified decomposable text, calculate the similarity of entity parallel relationships, and extract entity hierarchical relationships to generate an ontology; finally, build a BIM rules and specifications knowledge graph based on the formed ontology framework, match the entities and their relationships extracted from the text to the ontology framework, and store them in the BIM rules and specifications knowledge graph; S222, Semantic decomposition of images: First, use OCR technology to extract text from images; then perform text preprocessing to remove irrelevant information from the text extracted from the image; then, develop an entity relationship extraction model based on the Transformer architecture language large model to determine whether the image contains BIM rules and specifications, automatically identify and extract entities and their relationships, calculate the similarity of entity parallel relationships, extract entity hierarchical relationships, and generate an ontology; finally, build a BIM rule and specification knowledge graph based on the formed ontology framework, match the entities and their relationships extracted from the image to the ontology framework, and store them in the BIM rule and specification knowledge graph; S223. Semantic decomposition of tables: First, perform text preprocessing to identify and split the table structure, list the table header, table title, and each row of data in the table separately, and remove irrelevant information text extracted from the table; then, develop an entity relationship extraction model based on the Transformer architecture language large model, combine the table header and table title content, analyze the logical relationship between each data item in the table, automatically identify and extract entities and their relationships, calculate the similarity of entity parallel relationships, and extract entity hierarchical relationships to generate an ontology; finally, build a BIM rule and specification knowledge graph based on the formed ontology framework, match the entities and their relationships extracted from the table to the ontology framework, and store them in the BIM rule and specification knowledge graph.

[0014] Furthermore, the specific implementation steps of S3 are as follows: S31. Knowledge fusion: Fusion of multi-source description information about the same entity or concept, and elimination of redundant information fragments through technical means to obtain a complete description of the entity; S32. Knowledge processing: First, knowledge reasoning is performed. Starting from the existing entity relationship data, entity attributes, and conceptual hierarchical relationships in the BIM rules and specifications knowledge graph, new associations between entities are established using logic-based reasoning and graph-based reasoning algorithms to expand and enrich BIM rules and specifications. Then, quality assessment is implemented to quantify the credibility of BIM rules and specifications, and the quality of the knowledge graph is ensured by discarding knowledge with low confidence. Specifically, consistency checks are performed to ensure that there are no logical conflicts in the integrated BIM rules and specifications. Finally, if necessary, manual review methods are combined to verify the BIM rules and specifications through accuracy verification, expert verification, or use of external reliable data sources to verify the BIM rules and specifications, implement integrity measurement, and evaluate the comprehensiveness of the information in the BIM rules and specifications.

[0015] Further, the specific implementation steps of S31 are as follows: S311, Knowledge Fusion Engine: First, develop a knowledge fusion engine. Experts in the field of BIM engineering will formulate empirical rules to resolve various conflicts that may exist between entities based on their BIM engineering experience and their understanding of BIM standards. S312. Implement knowledge disambiguation: Use knowledge disambiguation technology to identify and resolve entities that may have different meanings in different contexts.

[0016] Further, the specific implementation steps of S4 are as follows: S41. Regularization of BIM rules and specifications: Regularize BIM rules and specifications, converting entities and their relationships from the BIM rules and specifications knowledge graph into regular expressions that can be directly integrated and imported into the BIM platform; S42. Form the natural language of BIM rules and specifications: First, extract the key elements in the entity recognition BIM rules and specifications, and then build a mapping to natural language for each identified rule element. Based on this mapping relationship, convert each BIM rule and specification clause into a natural language that people can understand; then, in the BIM engineering project design process, test the usability of the BIM rules and specifications in natural language form, use feedback based on the test results, adjust and optimize the mapping method, and finally obtain the natural language entries of the BIM rules and specifications to assist BIM forward design.

[0017] By adopting the above technical solution, the present invention has the following beneficial effects: 1) The present invention greatly reduces the time cost and error possibility of manual reading and collating of rule documents by efficiently, quickly and accurately extracting BIM rules and specifications from BIM standards, solves the heavy workload of manual reading, understanding and executing BIM standards, improves the efficiency of BIM in actual engineering projects and the efficiency of implementing the BIM standard system, and improves the accuracy of BIM rules and specifications in actual engineering projects.

[0018] 2) The text paragraph extraction technology of PDF documents proposed in this invention provides an efficient and accurate solution for extracting text paragraphs from PDF documents. Through sophisticated layout analysis and intelligent text reconstruction technology, it ensures that the text paragraphs extracted from the PDF document are not only complete in content, but also able to retain the original layout and format of the document.

[0019] 3) The table content extraction technology for PDF documents proposed in the present invention has unique advantages over the current tools and methods on the market in terms of processing the continuity of cross-page tables and table border recognition. It solves the problem of accurate reconstruction of cross-page tables and can output the table content in PDF documents in the form of structured data, which is convenient for further data analysis and application. Based on the proposed table intelligent parsing algorithm, not only the integrity, accuracy and effectiveness of table extraction are improved, but also the accurate reconstruction of the table structure is achieved, which greatly enhances the availability and operability of information. The present invention has demonstrated excellent performance and broad application potential in processing table extraction in PDF documents.

[0020] 4) The semantic decomposition technology of BIM standard document content proposed in the present invention, combined with the prompt word optimization technology and management method adopted in the present invention, can respectively implement semantic decomposition of three different forms of document content: text paragraphs, pictures, and tables, accurately extract entities and their relationships from various types of content with high professional complexity, and transform rule specifications into structured data that is easy to manage and apply. It can solve the problem of semantic understanding of BIM standards, greatly reduce the professional threshold of BIM standards, and greatly reduce the loss of manpower and material resources in implementing BIM engineering projects. 5) The BIM rules and specifications fusion and processing technology proposed in this invention solves the conflicts, unclear references and inconsistencies among BIM rules and specifications, and integrates fragmented BIM rules and specifications to form a comprehensive, three-dimensional and high-quality BIM rules and specifications system.

[0021] 6) The BIM rule specification regularization and semanticization technology proposed in the present invention, on the one hand, opens up the system integration of BIM rule specifications into the BIM platform, realizes the reverse review and verification of geometric depth and information depth in the BIM delivery process, and ensures the consistency and compliance of BIM project design, construction and operation and maintenance. On the other hand, it forms BIM rule specifications expressed in natural language, which greatly facilitates the forward design work of BIM engineers in BIM projects.

[0022] 7) The present invention realizes the automatic semanticization of BIM standards, which is the cutting-edge application of the latest achievements in the field of artificial intelligence in BIM technology, and realizes the deep integration of large models and BIM technology. The present invention reduces the dependence of engineering projects on professional BIM technicians, allowing even team members who do not have deep BIM background knowledge to participate in the review and application of rules, thereby improving the flexibility and adaptability of the project. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 A flowchart of the BIM rule specification extraction method based on a large model proposed in the present invention; Figure 2 This is a screenshot of a cross-page table of the BIM standard in an embodiment of the present invention; Figure 3 Identify the extracted table content in the embodiment of the present invention; Figure 4 This is a text screenshot of the BIM standard in an embodiment of the present invention; Figure 5 is a set of regularized BIM rule specifications in an embodiment of the present invention; Figure 6 It is a set of natural language BIM rule specifications in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0027] Combination Figure 1 As shown, an embodiment of the present invention provides a BIM rule specification extraction method based on a large model, which specifically includes the following steps: S1. Extraction of BIM standard document content The extraction of BIM standard document content is the basis for the extraction of BIM rules and specifications based on the big model, and is the cornerstone of the entire process. The efficiency and accuracy of the extraction of BIM standard document content directly determines the efficiency and accuracy of calling the big model for identification processing in the subsequent steps. The information in the BIM standard document has a complex structure and various forms, such as text descriptions, tables, and pictures, which brings challenges to calling the big model to semantically understand the BIM standard and extract BIM rules and specifications. Therefore, at the beginning of the entire process of extracting BIM rules and specifications based on the big model, it is necessary to first complete the extraction of BIM standard document content to provide accurate input for the semantic understanding of the big model and the extraction of rules and specifications.

[0028] The purpose of this step is to extract all the contents in the BIM standard document, including text paragraphs, pictures and tables, and process them accordingly, so as to convert them into a form that can be used for subsequent semantic understanding of the big model and extraction of BIM rules and specifications. This process mainly includes the preprocessing of standard documents, classification, identification and extraction of document content. Through preliminary content extraction, the foundation is laid for the extraction of BIM rules and specifications based on the big model.

[0029] S11. Convert BIM standard document format to PDF format PDF is an inherently structured document format that ensures that information is stored in a fixed layout and element arrangement, maintaining the visual layout and content of the file unchanged, which makes it more feasible and efficient to accurately identify and extract text, graphics and tables from PDF files. BIM standard documents are usually officially released in PDF format, but there are also BIM standard documents in other formats (such as doc or docx).

[0030] In this embodiment, in order to ensure the consistency of the document format and the smooth progress of the extraction process, the BIM standard document format must first be standardized and unified into the PDF format. If the BIM standard to be extracted is a PDF document, no processing is required, otherwise it should be converted into a PDF BIM standard document.

[0031] There are many ways or tools available for document format conversion, such as online file conversion services provided by SmallPDF, or format conversion functions provided by desktop applications such as Adobe Acrobat and WPS. In this embodiment, a programming method is used for conversion, and pypandoc is used to convert BIM standards other than PDF format into BIM standard documents in PDF format. pypandoc is a Python library based on Pandoc function implementation, which provides an interface for using Pandoc functions in the Python environment, and can process documents in various formats such as Markdown, HTML, Word docx, LaTeX, etc., and convert them into PDF format.

[0032] S12. Classification, recognition and extraction of PDF document content After step S11, the BIM standard document to be read has been unified into a PDF format document. This step is to extract the content of the PDF format document, which aims to extract all the content from the PDF format document, including three types of content: text paragraphs, pictures and tables, and perform corresponding processing to convert it into a form that can be used for subsequent semantic understanding and extraction of BIM rules and specifications.

[0033] Specifically, for the extraction of document content in PDF format, this embodiment develops a text extraction technology for PDF documents and a table content extraction technology for PDF documents, and the extraction of images is completed using an existing third-party library.

[0034] S121. Extraction of text paragraphs For the extraction of text paragraphs, currently commonly used methods and tools include PDF document reading and editing tools Adobe Acrobat, third-party library PDFMiner, third-party library PyMuPDF, etc. They can extract the text paragraph content of PDF under certain conditions, but these methods and tools have many limitations. First, some tools have difficulty accurately identifying and maintaining the original document structure when dealing with complex layouts and multi-column documents, resulting in the lack of logic and coherence in the extracted text. In addition, for documents containing multiple font sizes and styles, these technologies often cannot effectively distinguish between titles, text, and footnotes, thus affecting the accuracy of subsequent content analysis. Another common problem is that when a document contains both text and images, many existing technologies have difficulty correctly distinguishing and processing the relationship between text and images, further reducing the quality of information extraction.

[0035] This embodiment develops a text extraction technology for PDF documents, which is used to accurately extract text paragraphs from PDF format documents. The technology is based on an in-depth analysis of the structure and layout characteristics of the PDF document, including text position, font size, and line spacing, to accurately reconstruct the text paragraphs in the document.

[0036] This technology first analyzes the internal structure of the PDF file to identify the text flow and layout attributes, including font size, line spacing, and text location information, focusing on the spatial distribution and alignment between text blocks. Then, using this layout information, this method can distinguish independent text blocks and organize them into paragraphs according to visual continuity. Specifically, it uses algorithms to identify the logical associations between texts, uses paragraph indents or list marks, font size changes to distinguish text paragraphs, and identifies titles and subtitles. This technology is suitable for processing PDF documents with rich formats and multiple elements. Considering the layout characteristics of the text, by implementing a highly optimized text merging process, text fragments in the same or adjacent lines are intelligently merged according to their relative positions in the document to form coherent text paragraphs.

[0037] The text paragraph extraction technology of PDF documents proposed in this step provides an efficient and accurate solution for extracting text paragraphs from PDF documents. Through sophisticated layout analysis and intelligent text reconstruction technology, it ensures that the text paragraphs extracted from PDF documents are not only complete in content, but also able to retain the original layout and format of the document.

[0038] S122. Extraction of table contents For the extraction of text paragraphs, currently commonly used tools include: open source tool Tabula, python library Camelot, third-party library PDFplumber, etc. They can extract the table content of PDF under certain conditions, but these tools have many limitations. First, these existing technologies extract tables based on a single page of the document. When dealing with cross-page tables or complex table structures containing multi-level titles, the original data integrity and logical relationships cannot be maintained, and the complete parsing and extraction of table semantics cannot be achieved. Secondly, when the table and the surrounding text are complex in layout or lack clear separation, the existing technology has difficulty in identifying the table boundaries and structure. In addition, the existing solutions have insufficient support for non-standard table styles. For example, the cell merging phenomenon in the table cannot be correctly identified and extracted, which leads to inaccurate or missing data extraction results.

[0039] This embodiment develops a table content extraction technology for PDF documents, which is used to effectively, completely and accurately extract table content from PDF format documents. This technology is based on the analysis of the relative layout structure of tables and pages in PDF documents and the analysis of table content. It can accurately identify and parse the data in the table when the document page layout is complex, and realize the automatic merging of cross-page tables and the automatic filling of merged cells.

[0040] This technology first analyzes the page elements of the PDF document to identify the basic lines and text blocks that constitute the table. By analyzing the geometric positions and relative relationships of these elements, the boundaries, row and column structures, and cell positions in the page of the table are determined. Then, based on the determined cell positions, the text content is extracted, and the alignment, font size, and style of the text are identified to accurately reconstruct the data in the cell. In particular, this technology has developed an advanced table intelligent parsing algorithm based on PDF documents, which is advanced in two aspects: first, the algorithm can realize the automatic merging of cross-page tables based on the relative positions of the table boundaries between different pages and the spatial layout of the table on the corresponding page, ensuring the integrity and accuracy of the cross-page table extraction results; second, the algorithm can correctly parse other special processing scenarios such as merged cells in the table based on the identified cell position coordinates, and realize the automatic filling of merged cell content based on the cell position coordinates and the table space layout, ensuring the effectiveness of the cross-page table extraction results. Based on this, this technology solves many difficulties faced by existing table extraction technologies and effectively improves the integrity, accuracy, and parsing efficiency of table content extraction.

[0041] Regarding the extraction of table content implemented in this step, here is Figure 2 , Figure 3 Take this as an example. Figure 2 The following is a screenshot from the JTG / T 2421-2021 Highway Engineering Design Information Model Application Standard document. This table is a cross-page table, the continuation table has no title and there are merged cells. The existing common technology cannot intelligently and automatically identify the tables on the two pages as a unified table, and the extracted results cannot reproduce the true structure of the table for the merged cell phenomenon. However, Figure 4 The technology shown can automatically extract the cross-page table as a whole and realize automatic filling of the content.

[0042] The table content extraction technology of PDF documents proposed in this step analyzes the page elements of PDF documents and proposes an advanced table intelligent parsing algorithm based on PDF documents, and finally realizes the output of table content in PDF documents in the form of structured data. It not only improves the integrity, accuracy and effectiveness of table extraction, but also realizes the precise reconstruction of table structure, greatly enhancing the availability and operability of information. The present invention has demonstrated excellent performance and wide application potential in processing table extraction in PDF documents.

[0043] S123. Image extraction Extracting images from PDF is a common requirement, and currently various commonly used technical means are sufficient to meet the demand. Common technical means for extracting images from PDF documents include: using PDF editing or reading software such as Adobe Acrobat, using online tools and services such as SmallPDF, and using programming libraries such as PDFMiner.

[0044] In the present embodiment, PDFMiner is used to complete the extraction of images in the document. PDFMiner is a third-party Python library that can identify various elements in a document by analyzing the internal object structure of a PDF file. For image extraction, PDFMiner first locates the image objects in the PDF, which are usually embedded in the page content stream of the PDF. Then, it decodes the compression algorithm used by these image objects, extracts the image data, and can save it as an external file. This process involves a deep understanding of the PDF file format and the processing capabilities of the image encoding format. PDFMiner provides the ability of underlying access to PDF files, allowing users to extract almost any type of information in a document on demand, including images embedded therein.

[0045] S2. Semantic deconstruction of BIM standard document content The BIM standard system covers various BIM rules and specifications from design to construction, and then to operation and maintenance, ensuring that all parties involved in the design, construction, and operation and maintenance stages can follow unified standards on BIM. However, the BIM-related standard systems in various industries are complex and contain a lot of content. At the same time, these standards are constantly being updated and improved. The workload of manually entering and maintaining these standards is huge, which brings challenges to the accurate understanding and rapid application of BIM standards. With the development of artificial intelligence technology, especially the emergence of large models, a new solution has been provided. By utilizing the powerful natural language processing capabilities of large models and implementing a complete prompt word project, the content extracted from BIM standard documents can be intelligently semanticized, the extraction of BIM rules and specifications can be completed, and then converted into structured data that is easy to manage and apply, thereby greatly improving work efficiency and supporting the rapid update and deployment of BIM standards.

[0046] BIM standards cover complex and multifaceted BIM rules and regulations, which can be divided into three categories: BIM naming / coding rules and regulations, BIM engineering object / model unit delivery requirements, and BIM engineering object / model unit geometry / information depth requirements. In addition, different chapters of BIM standards and different BIM standards may have certain relationships, such as reference or inheritance relationships, so it is also necessary to identify BIM standard reference relationships.

[0047] In this embodiment, step S1 completes the extraction of text, pictures and tables in the standard document. This step realizes the semantic understanding and automatic semanticization of BIM rules and specifications based on the big model, that is, based on the content extracted in step S1, the entities and their relationships are extracted for use in subsequent steps to form BIM rules and specifications. Specifically, it includes the establishment and management of prompt word libraries, semantic decomposition of texts, semantic decomposition of pictures and semantic decomposition of tables. The specific steps are as follows: S21. Prompt word library and its management A prompt is a text or instruction that provides input to a large model to guide it to generate a specific output. It is a series of precise instructions or questions designed to guide a large model to perform a specific task. When generating text, the large model will try to understand the prompt and generate a corresponding response based on its understanding. The prompt is the key to interacting with the large model. It not only defines the type of task that the user wants the model to perform, but also directly affects the quality and relevance of the model's output. The results returned by the large model are usually text strings. In order to better guide the large model to complete the expected task, the prompt needs to be carefully optimized.

[0048] The optimization methods of prompt words mainly include: expressing instructions clearly, such as including more details in the query to improve the relevance of the answer, clarifying that the large model adopts a specific role perspective when answering, using separators to clearly divide the different parts of the prompt word, providing examples, and specifying the expected output length; providing reference text, such as instructing the model to use the provided reference text to answer, instructing the model to use quotations in the reference text to answer; breaking down complex tasks into simpler subtasks, such as using intent classification to identify the instructions most relevant to the user's query, for applications involving long conversations, aggregating or filtering previous conversations, and recursively constructing complete summaries for long documents with segmented summaries; using external tools, such as embedding-based search to achieve efficient knowledge retrieval, using code execution for more precise calculations or calling external APIs.

[0049] In this embodiment, in order to ensure that stable and easy-to-process return results can be obtained when calling the large model, and to achieve more advanced processing and application, a clear structured format should be specified in the prompt word to ensure that the return results can be processed, point out clear task requirements to avoid misunderstanding of the large model, and use the thinking chain prompt technology to guide the large model to perform better reasoning ability. The specific optimization steps are as follows: S211. Specify the format of the result returned by the large model. First, a requirement is put forward in the prompt word, stipulating that the large model returns a clear structured format. In this embodiment, the large model is required to return the result in JSON format. JSON (JavaScript Object Notation) is a lightweight data exchange format that stores and transmits attribute-value pairs and array data types in text form. JSON is easy to read and write, and is also easy to machine parse and generate. It is widely used for data transmission in network applications. Clearly proposing to use the return result in JSON format can avoid the uncertainty of the return result of the large model and is easy to handle.

[0050] S212, point out the clear semantic decomposition task. Then, clearly point out in the prompt that the task of this dialogue is to "realize the semantic decomposition of the content extracted from the BIM standard document, and the returned results will be used to extract and form BIM rules and regulations."

[0051] S213, chain of thought prompts. Then, the chain of thought prompt technology should be provided in the prompt words, and the thinking process of thinking about the problem and reasoning and answering of the large model should be described in natural language, and finally the correct result of the example problem is guided. For the different types of content extracted from the standard document in step S1, the types of BIM rules and specifications contained therein are also different, and the chain of thought prompts that are in line with the specific actual implementation are required.

[0052] S214, separation and definition of prompt words. Prompt words are a block of text input into the large model, and may be accompanied by some other multimodal input information. In order to ensure that the large model can better understand the prompt words and avoid ambiguity, separators should be used to separate and define different parts of the prompt words. Commonly used separators include triple quotation marks, XML tags, special symbols such as book title marks. In this embodiment, triple quotation marks are used to separate and define different parts of the prompt words. Specifically, the format requirement text of the result returned by the large model, the text of the proposed semantic decomposition task, the thinking chain prompt text, the sample text, and the text content to be semantically decomposed are respectively marked with triple quotation marks to ensure that the large model correctly understands the structure and composition of the prompt words.

[0053] The following is an example of the "prompt word library and its management" implemented in this step. After the optimization of the prompt word technology in step S21, the corresponding prompt words are finally formed as follows: Text paragraph: “““ SENTENCES ””” Require: 1. The task is: the above text paragraph is extracted from the BIM standard document. Based on the given text paragraph input, semantic decomposition is implemented to extract the entities and their relationships in the text paragraph. The returned results will be used to extract and form BIM rules and regulations.

[0054] 2. Format requirement: The returned result is in JSON format.

[0055] For example: To help you better understand the task objectives, thought process, and return better results, the following examples are given.

[0056] 1) Example of a text paragraph: "The building floor height is above 2.20 meters (including 2.20 meters)".

[0057] 2) Thinking chain: Entity recognition: Identify "house height" and "2.20 meters" as key entities.

[0058] Relationship extraction: Extract "the building floor height should be above 2.20 meters (including 2.20 meters)" as the relationship.

[0059] Intent understanding: Understand that the text requires that the floor height of the house should not be less than 2.20 meters.

[0060] 3. Return results: “““ { "entities": [ { "entity": "House Height", "start_index": 0, "end_index": 4 }, { "entity": "2.20 meters", "start_index": 6, "end_index": 11 } ], "relations": [ { "relation": "The building floor height should be above 2.20 meters (including 2.20 meters)", "start_index": 0, "end_index": 11 } ] } ””” Based on the requirements I have put forward and referring to the examples provided, please implement semantic decomposition for the given text paragraph input and return an answer that meets the requirements.

[0061] This prompt word combines all the optimization techniques proposed in step S21 and is an example of a prompt word template for the semantic decomposition task of text in this embodiment. In specific use, the text content extracted from the BIM standard document can be replaced with SENTENCES in the prompt word template.

[0062] S22. Semantic Deconstruction The BIM standard covers complex and multifaceted BIM rules and regulations, which can be mainly divided into three categories: BIM naming / coding rules and regulations, BIM engineering object / model unit delivery requirements, and BIM engineering object / model unit geometry / information depth requirements. The three types of BIM rules and regulations exist in different ways in the BIM standard document. Therefore, for the content extracted in step S1, the types of BIM rules and regulations contained in different forms of content are also different.

[0063] Semantic decomposition is to extract entities and their relationships from data sources to obtain candidate knowledge units, and then further transform entities and their relationships into rules and regulations. Ontology refers to a formal naming and definition of the types, attributes, and relationships of entities in a specific field or at all. It refers to a recognized set of concepts and conceptual frameworks, such as "people", "things", "objects", etc., which is actually a formal expression of a set of concepts and their relationships in a specific field. The key issue of semantic decomposition is how to automatically extract information from heterogeneous data sources, which usually requires the use of natural language processing and other technologies to extract structured information.

[0064] In this step, different semantic decomposition techniques are used for the three types of content, namely text, pictures and tables, extracted from the standard document in step S1 to extract the entities and their relationships in the response content for use in subsequent steps to form BIM rules and specifications.

[0065] S221. Semantic deconstruction of text The text of BIM standard documents has certain structural characteristics and contains a large number of professional terms. The professionalism and complexity of BIM standards make semantic decomposition more difficult. In the process of semantic decomposition of the text of BIM standard documents, the key is to accurately parse the complex terms and instructions, which involves detailed identification and classification of each element in the technical text, such as term definitions, performance standards, and operating procedures. Developing targeted algorithms and models that can understand and process these professional contents is the key to successfully achieving semantic decomposition.

[0066] In this embodiment, the specific steps of semantic decomposition of text are as follows.

[0067] First, the text is preprocessed to clean the text and remove noise such as format errors or irrelevant information such as headers and footers. Then, a text classification algorithm is developed based on the BERT large model to realize the analysis and evaluation of the decomposability of the text. The BERT model has a strong deep understanding ability in natural language processing. It can accurately classify whether the text has decomposable structural features through a comprehensive understanding of the context. In this way, it can realize the recognition and classification of rule-based text. Then, based on the results of the preliminary classification, an entity relationship extraction model based on the Transformer architecture language large model is developed to automatically identify and extract entities and their relationships from the classified decomposable text, calculate the similarity of entity parallel relationships, and extract entity superior and subordinate relationships to generate ontologies. These ontology frameworks are the key to building knowledge graphs and forming BIM rules and specifications. Finally, a BIM rule and specification knowledge graph is constructed based on the formed ontology framework, and the entities and their relationships extracted from the text are matched to the ontology framework and stored in the BIM rule and specification knowledge graph.

[0068] The following is an example of the semantic decomposition of the text achieved in this step. Figure 4 As shown in the figure, the content is Article 4.0.4.1 of the "CJJ 11-2011 Urban Bridge Design Code". This technology identifies and extracts entities such as "bridge pier", "axis", "skew", "intersection angle", and "clear width of navigation hole". The extracted relationships are "the axis of the bridge pier along the water flow direction should be consistent with the mainstream direction of the highest navigation water level", "when the skew is crossed, the intersection angle should not be greater than 5°", "when the intersection angle is greater than 5°, the clear width of the navigation hole should be increased", etc., and the above information content is stored in the BIM rule specification knowledge graph.

[0069] Through the above semantic decomposition process, the semantic decomposition of the text in the BIM standard document is realized, and the entities and their relationships in the text are extracted.

[0070] S222. Semantic decomposition of images The pictures of BIM standard documents may be diagrams of naming / coding forms, which are used to illustrate the BIM naming / coding rules and regulations. The arrangement of text content in such pictures is more structural and more conducive to semantic decomposition. In the semantic decomposition of pictures, in addition to the analysis and extraction of entities and their relationships, OCR technology is also involved. OCR (optical character recognition) technology is used to recognize and convert printed or handwritten text from images or scanned documents into machine-readable data. It is widely used in data entry, file digitization and information automation processing.

[0071] In this embodiment, the specific steps of semantic decomposition of the image are as follows.

[0072] First, OCR technology is used to extract text from images. Then, text preprocessing is performed to remove irrelevant information from the text extracted from the image. Then, an entity relationship extraction model based on the Transformer architecture language large model is developed to determine whether the image contains BIM rules and specifications, automatically identify and extract entities and their relationships, calculate the similarity of entity parallel relationships, and extract entity hierarchical relationships to generate an ontology. Finally, a BIM rules and specifications knowledge graph is constructed based on the formed ontology framework, and the entities and their relationships extracted from the image are matched to the ontology framework and stored in the BIM rules and specifications knowledge graph.

[0073] Through the above semantic decomposition process, the semantic decomposition of pictures in BIM standard documents is realized, and the entities and their relationships in the pictures are extracted.

[0074] S223. Semantic decomposition of tables The tables in BIM standard documents may include the delivery requirements of BIM engineering objects / model units and the geometric / information depth requirements of BIM engineering objects / model units. These rules and regulations stipulate the requirements for geometric expression accuracy and information depth at different depths during BIM delivery, guide the refinement of BIM components and the amount of information required, and directly affect the practicality of BIM and its support for later operation and maintenance. The table content of BIM standard documents is highly structured. Compared with text and pictures, there is no need to spend too much energy on semantic analysis. The focus is on understanding the table content and processing the data after analysis and understanding.

[0075] In this embodiment, the specific steps of semantic decomposition of the table are as follows.

[0076] First, text preprocessing is performed to identify and split the table structure, list the table header, table title, and each row of data in the table, and remove irrelevant information text extracted from the table. Next, an entity relationship extraction model based on the Transformer architecture language large model is developed. Combined with the table header, table title and other contents, the logical relationship between each data item in the table is analyzed, and entities and their relationships are automatically identified and extracted. The similarity of parallel entity relationships is calculated, and the entity hierarchical relationship is extracted to generate an ontology. Finally, a BIM rule and specification knowledge graph is constructed based on the formed ontology framework, and the entities and their relationships extracted from the table are matched to the ontology framework and stored in the BIM rule and specification knowledge graph.

[0077] The following is an example of the semantic decomposition of the text achieved in this step. Figure 2 The table shown and Figure 3 As shown in the extracted table content, this technology identifies the table title, table header, and each row of data in the table, removes the irrelevant information text of the note column, and respectively identifies and extracts entities such as "longitudinal section straight line", "identification code", "classification code", and "starting pile number", and forms relationships in combination with the requirements of information depth L2.0 and L3.0. For example, for the "starting pile number", a relationship is formed such as "the starting pile number should be available at the information depth L2.0" and "the starting pile number should be available at the information depth L3.0", and the above information content is stored in the BIM rule and specification knowledge graph.

[0078] Through the above semantic decomposition process, the semantic decomposition of the table in the BIM standard document is realized, and the entities and their relationships in the table are extracted.

[0079] S3. Integration and processing of BIM rules and regulations In the previous steps, we have extracted various forms of content from the BIM standard document and extracted the corresponding entities and their relationships. The task of this step is to form BIM rules and specifications based on the extracted entities and their relationships through knowledge fusion and knowledge processing.

[0080] In this embodiment, knowledge fusion is first performed on the entities and their relationships extracted in step S2 through knowledge disambiguation and other technologies, and multi-source description information about the same entity or concept is integrated to form a complete three-dimensional entity; then knowledge processing is completed through knowledge reasoning and other technologies to form BIM rules and specifications. The specific steps are as follows: S31. Knowledge Fusion Knowledge fusion is a strategy to integrate information from different sources into a unified and coordinated knowledge base. This includes not only adding new information to the existing structure, but also resolving conflicts and inconsistencies between data. Knowledge disambiguation is the process of identifying and distinguishing multiple entities with the same name or expression in different contexts. It is a key technology for identifying and resolving ambiguity problems in the knowledge fusion process, solving the problem of synonyms and the relationship between different entities, and ensuring that the same name or term is accurately distinguished in different contexts.

[0081] For the content of BIM standard documents, the entities and their relationships obtained after step S2 are flat, lacking hierarchy and logic. Different knowledge focuses on different aspects, and there may be a large number of redundant information fragments. Knowledge fusion is to integrate multi-source description information about the same entity or concept, and eliminate redundant information fragments through technical means such as knowledge disambiguation to obtain a complete description of the entity.

[0082] In this embodiment, the specific steps of knowledge fusion are as follows.

[0083] S311, knowledge fusion engine. First, develop a knowledge fusion engine. Experts in the field of BIM engineering, based on their experience in BIM engineering and their understanding of BIM standards, formulate empirical rules to resolve various possible conflicts between entities. For example, for the same entity, if there is a conflict between the rules of different clauses, the latest release time of the BIM standard from which the entity comes shall prevail. The empirical rules set in this embodiment include but are not limited to the above examples. In order to cover as many possible conflict situations as possible, continuous feedback optimization and improvement are required in specific applications.

[0084] S312. Implement knowledge disambiguation. Use knowledge disambiguation technology to identify and resolve entities that may have different meanings in different contexts. For example, in a BIM project for bridges, a "beam" may be a horizontal beam or a longitudinal beam of an arch bridge, or a bridge tower tie beam of a cable-stayed bridge. For these different meanings of "beams", the BIM standard has corresponding different regulations.

[0085] Specifically, it includes the following technologies: based on natural language processing technology, the true meaning of entities is examined through context analysis; entity linking, the extracted entities and their relationships are linked to the correct entity objects in the knowledge graph, and a group of candidate entity objects are selected from the knowledge graph based on the given entity reference items, and then the reference items are linked to the correct entity objects through similarity calculation; cluster analysis is performed to achieve grouping based on similar features and distinguish entities with similar names in different contexts.

[0086] S32. Knowledge Processing The above steps have completed the extraction of various forms of content from BIM standard documents, extracted the corresponding entities and their relationships, and achieved knowledge fusion to ensure that the same name or term is accurately distinguished in different contexts, and obtained a series of factual expressions about BIM rules and regulations. In order to further improve the quality of the extracted BIM rules and regulations, knowledge processing is also required to incorporate qualified parts into the BIM rules and regulations system to ensure its quality.

[0087] In this embodiment, the specific steps of knowledge processing are as follows. First, knowledge reasoning is performed. Starting from the existing entity relationship data, entity attributes, and conceptual hierarchical relationships of entities in the BIM rules and specifications knowledge graph, logic-based reasoning and graph-based reasoning algorithms are used to establish new associations between entities, expand and enrich BIM rules and specifications. Then, a quality assessment is implemented to quantify the credibility of BIM rules and specifications, and the quality of the knowledge graph is guaranteed by discarding knowledge with low confidence. Specifically, a consistency check is performed to ensure that there are no logical conflicts in the integrated BIM rules and specifications. Finally, if necessary, the manual review method is combined to verify the BIM rules and specifications through accuracy verification, expert verification, or the use of external reliable data sources, and integrity measurement is implemented to evaluate the comprehensiveness of the information in the BIM rules and specifications.

[0088] This step completes the knowledge processing of the BIM rules and specifications knowledge graph, ensuring the quality of BIM rules and specifications.

[0089] S4. Regularization and semanticization of BIM rules and regulations On the one hand, BIM rules and specifications can be used to achieve reverse review and verification of geometric depth and information depth in the BIM delivery process, which involves the system integration of BIM rules and specifications into the BIM platform; on the other hand, from the BIM design perspective, it can also achieve positive assistance in the BIM design process, therefore involving the natural language expression of rules and specifications from a user perspective.

[0090] In this embodiment, the above steps realize the transformation from BIM standard documents to a comprehensive and three-dimensional BIM rules and regulations system in the BIM rules and regulations entity map. This step mainly realizes the system integration of BIM rules and regulations from the BIM rules and regulations knowledge map to the BIM platform through BIM rules and regulations regularization, and forms the transformation of the user-oriented BIM rules and regulations natural language. The specific steps are as follows: S41, BIM rules and regulations regularization This step mainly regularizes the BIM rules and specifications, converting the entities and their relationships from the BIM rules and specifications knowledge graph into regular expressions that can be directly integrated and imported into the BIM platform.

[0091] In BIM software, rules and specifications usually exist in the form of standardized codes or expressions, which can be a series of parametric commands or data structures defined in the IFC (Industry Foundation Classes) standard. These rules and specifications not only contain definitions of BIM attributes, but also logic on how to perform model verification, such as "there is a civil engineering element A whose category matches "1" and all civil engineering element A's size information | skew angle regular "^[0-9]+$"". For example, they may include regular expressions to check whether the naming of model elements conforms to a specific format, or logical rules to verify whether a part of the model meets specific design parameters. In order to enable BIM rules and specifications to be applied in BIM software, the regularization process may include converting naming rules, delivery requirements, geometry and information depth requirements, etc. into IFC standard entities and relationships, or into verification scripts that can be recognized by the software.

[0092] In this embodiment, the specific steps of regularization of BIM rules and specifications are as follows: first, extract key elements in entity recognition BIM specification rules, such as naming rules, parameter requirements, etc., and then build a mapping for each identified rule element. This mapping converts the entity and its relationship description into a regular expression or code structure that can be understood by the BIM software; then, according to the mapping, write a regular expression for each BIM rule to ensure that they can accurately match or verify the relevant data in the BIM model; then, test the written regular expressions in the BIM software environment to ensure that they can correctly perform the expected verification; finally, according to the test results, adjust and optimize the regular expressions to obtain a set of regular expressions for BIM rule specifications, which is the final set of BIM rule specifications.

[0093] The following is an example of the "BIM rule regularization" implemented in this step. Figure 2 The table shown and Figure 3 The extracted table content shown in the figure, for the content of "starting pile number", forms a regularized BIM rule specification such as "there is a civil engineering element A whose category matches "longitudinal section straight line" and the location information of all civil engineering elements A|starting pile number is regular "^[0-9]+$"".

[0094] In this embodiment, the final set of regularized BIM rule specifications obtained in this step is stored in an Excel file and imported into the BIM software for reverse audit verification of BIM delivery, such as Figure 5 shown.

[0095] S42. Form a natural language for BIM rules and regulations In addition to being in the form of regular expressions, which are used for reverse review and verification of geometric depth and information depth during the BIM delivery process, BIM rule specifications also need to be in the form of natural language that is easy for people to understand, to implement the auxiliary process of forward design of BIM projects.

[0096] In this embodiment, the specific steps for forming the natural language of BIM rules and specifications are as follows: first, extract the key elements in the entity recognition BIM rules and specifications, then build a mapping to natural language for each identified rule element, and convert each BIM rule and specification clause into a natural language that people can understand based on this mapping relationship; then, in the BIM engineering project design process, test the usability of the BIM rules and specifications in natural language form, use feedback based on the test results, adjust and optimize the mapping method, and finally obtain the natural language entries of the BIM rules and specifications to assist BIM forward design.

[0097] The following is an example of how to achieve the "formation of BIM rule specification natural language" in this step. Figure 2 The table shown and Figure 3 The extracted table content shown in the figure, for the content of "starting pile number", forms the BIM rule specification natural language such as "the longitudinal section straight line should have the position information | starting pile number at the information depth L2.0", "the longitudinal section straight line should have the position information | starting pile number at the information depth L3.0".

[0098] In this embodiment, the set of BIM rule specification natural language obtained in this step is stored in an Excel file and imported into the BIM software for forward assistance in the BIM design process, such as Figure 6 shown.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A BIM rule specification extraction method based on a large model, characterized in that: The steps include: S1. Extraction of BIM standard document content: Extract all the contents of the BIM standard document and process them accordingly, so as to transform them into a form that can be used for subsequent semantic understanding of the large model and extraction of BIM rules and specifications; S2. Semantic decomposition of BIM standard document content: Based on the content extracted in step S1, extract the entities and their relationships therein for use in subsequent steps to form BIM rules and specifications; S3. Integration and processing of BIM rules and specifications: Based on the extracted entities and their relationships, BIM rules and specifications are formed through knowledge integration and knowledge processing; S4. Regularization and semanticization of BIM rules and specifications: Through the regularization of BIM rules and specifications, the system integration of BIM rules and specifications from the BIM rules and specifications knowledge graph to the BIM platform is realized, and the user-oriented natural language conversion of BIM rules and specifications is formed.

2. The BIM rule specification extraction method based on a large model according to claim 1 is characterized in that: The specific implementation steps of S1 are as follows: S11. Convert BIM standard document format to PDF format; S12. Classification, recognition and extraction of document content in PDF format.

3. The BIM rule specification extraction method based on a large model according to claim 2 is characterized in that: The specific implementation steps of S12 are as follows: S121. Extraction of text paragraphs: First, by parsing the internal structure of the PDF file, the text flow and layout attributes are identified, including font size, line spacing, and text position information, focusing on the spatial distribution and alignment between text blocks; then, the layout information is used to distinguish independent text blocks and organize them into paragraphs according to visual continuity; S122, extracting table content: first, by analyzing the page elements of the PDF document, the basic lines and text blocks constituting the table are identified, and by analyzing the geometric positions and relative relationships of these elements, the table boundaries, row and column structures, and the positions of cells in the page are determined; then, based on the determined cell positions, the text content is extracted, and the alignment, font size, and style information of the text are identified to accurately reconstruct the data in the cell; S123. Image extraction: PDFMiner is used to extract images from documents.

4. The method for extracting BIM rules and specifications based on a large model according to claim 1, characterized in that: The specific implementation steps of S2 are as follows: S21. Prompt word library and its management: to ensure that stable and easy-to-process return results can be obtained when calling large models, and to achieve more advanced processing and application; S22, semantic decomposition: By extracting entities and their relationships from the data source, candidate knowledge units are obtained, and then the transformation from entities and their relationships to rules and regulations is further realized in the subsequent process.

5. The method for extracting BIM rules and specifications based on a large model according to claim 4 is characterized in that: The specific implementation steps of S21 are as follows: S211. Specify the format of the result returned by the large model; S212, point out clear semantic disassembly tasks; S213, thinking chain prompts; S214, the prompt words are separated and defined.

6. The method for extracting BIM rules and specifications based on a large model according to claim 4, characterized in that: The specific implementation steps of S22 are as follows: S221. Semantic decomposition of text: First, perform text preprocessing to clean the text and remove noise such as format errors or irrelevant information such as headers and footers. Then, develop a text classification algorithm based on the BERT large model to achieve decomposability analysis and evaluation of the text. Then, based on the results of the preliminary classification, an entity relationship extraction model based on the Transformer architecture language large model was developed to automatically identify and extract entities and their relationships from the classified decomposable text, calculate the similarity of entity parallel relationships, and extract entity hierarchical relationships to generate an ontology; finally, a BIM rule and specification knowledge graph was constructed based on the formed ontology framework, and the entities and their relationships extracted from the text were matched to the ontology framework and stored in the BIM rule and specification knowledge graph; S222, Semantic decomposition of images: First, use OCR technology to extract text from images; then perform text preprocessing to remove irrelevant information from the text extracted from the image; then, develop an entity relationship extraction model based on the Transformer architecture language large model to determine whether the image contains BIM rules and specifications, automatically identify and extract entities and their relationships, calculate the similarity of entity parallel relationships, extract entity hierarchical relationships, and generate an ontology; finally, build a BIM rule and specification knowledge graph based on the formed ontology framework, match the entities and their relationships extracted from the image to the ontology framework, and store them in the BIM rule and specification knowledge graph; S223. Semantic decomposition of tables: First, perform text preprocessing to identify and split the table structure, list the table header, table title, and each row of data in the table separately, and remove irrelevant information text extracted from the table; then, develop an entity relationship extraction model based on the Transformer architecture language large model, combine the table header and table title content, analyze the logical relationship between each data item in the table, automatically identify and extract entities and their relationships, calculate the similarity of entity parallel relationships, and extract entity hierarchical relationships to generate an ontology; finally, build a BIM rule and specification knowledge graph based on the formed ontology framework, match the entities and their relationships extracted from the table to the ontology framework, and store them in the BIM rule and specification knowledge graph.

7. The method for extracting BIM rules and specifications based on a large model according to claim 1, characterized in that: The specific implementation steps of S3 are as follows: S31. Knowledge fusion: Fusion of multi-source description information about the same entity or concept, and elimination of redundant information fragments through technical means to obtain a complete description of the entity; S32. Knowledge processing: First, knowledge reasoning is performed. Starting from the existing entity relationship data, entity attributes, and conceptual hierarchical relationships in the BIM rules and specifications knowledge graph, logic-based reasoning and graph-based reasoning algorithms are used to establish new associations between entities, expand and enrich BIM rules and specifications; then, quality assessment is implemented to quantify the credibility of BIM rules and specifications, and the quality of the knowledge graph is guaranteed by discarding knowledge with low confidence. Specifically, consistency checks are performed to ensure that there are no logical conflicts in the integrated BIM rules and specifications; finally, if necessary, manual review methods are combined to verify the BIM rules and specifications through accuracy verification, expert verification, or use of external reliable data sources to verify the BIM rules and specifications, implement integrity measurement, and evaluate the comprehensiveness of the information in the BIM rules and specifications.

8. The method for extracting BIM rules and specifications based on a large model according to claim 7, characterized in that: The specific implementation steps of S31 are as follows: S311, Knowledge Fusion Engine: First, develop a knowledge fusion engine. Experts in the field of BIM engineering will formulate empirical rules to resolve various conflicts that may exist between entities based on their BIM engineering experience and their understanding of BIM standards. S312. Implement knowledge disambiguation: Use knowledge disambiguation technology to identify and resolve entities that may have different meanings in different contexts.

9. The method for extracting BIM rules and specifications based on a large model according to claim 1, characterized in that: The specific implementation steps of S4 are as follows: S41. Regularization of BIM rules and specifications: Regularize BIM rules and specifications, converting entities and their relationships from the BIM rules and specifications knowledge graph into regular expressions that can be directly integrated and imported into the BIM platform; S42. Form the natural language of BIM rules and regulations: First, extract and identify the key elements in the BIM rules and regulations, and then build a mapping to the natural language for each identified rule element. Based on this mapping relationship, convert each BIM rule and regulation clause into a natural language that people can understand; then, in the design process of the BIM engineering project, test the usability of the BIM rules and regulations in natural language form, use feedback based on the test results, adjust and optimize the mapping method, and finally obtain the natural language entries of the BIM rules and regulations to assist the BIM forward design.

Images