Building information model automation compliance examination method driven by large language model

Abstract

The invention relates to the technical field of intellectualization, in particular to a building information model automation compliance review method driven by a large language model. According to the method, through the steps of efficient entity recognition and preliminary verification, precise IFC class mapping, efficient attribute information extraction and text conversion, automatic compliance review and the like, automatic recognition and reasoning of complex laws and regulations are achieved, and meanwhile efficient and accurate alignment between a BIM model and a standard text is achieved. According to the technical scheme, the automation degree and accuracy of BIM compliance review are greatly improved, the cost and risk of manual review are reduced, and powerful support is provided for digital transformation and intelligent management of the building industry.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent technology, and in particular relates to an automated compliance review method for a building information model driven by a large language model. Background Art

[0002] With the digital transformation of the construction industry, Building Information Modeling (BIM) technology has become a key tool for improving the efficiency of design, construction, and operation and maintenance of construction projects due to its powerful information integration and collaborative design capabilities. BIM models not only contain rich geometric information, but also cover a full range of building element properties such as building materials, structures, and equipment, which provides unprecedented data support for the compliance review of construction projects. However, faced with complex and ever-changing building design specifications and huge BIM model data, traditional manual review methods seem to be unable to cope with the time-consuming and error-prone characteristics that make it difficult to meet the efficient and accurate management needs of modern construction projects.

[0003] At present, although there are some automated BIM compliance review methods based on rules and shallow semantic parsing, these methods are stretched when dealing with complex regulatory logic and implicit rules. The root of the problem is that these methods often rely on preset rule bases for matching, lack an understanding of the deep semantics of the specification provisions, and find it difficult to accurately capture the logical relationship between the provisions. At the same time, the alignment between the BIM model and the specification text is also a major problem. Existing technologies can often only achieve simple attribute matching, but cannot achieve deep associations based on context and semantics. In addition, due to the diversity and regional differences in building specifications, the maintenance and update costs of the rule base are high, which limits the widespread application of automated review methods.

[0004] The difficulty of solving the problem is mainly reflected in two aspects: first, how to build a large language model that can accurately understand the deep semantics of the specification and realize the automatic recognition and reasoning of complex regulatory logic; second, how to achieve efficient and accurate alignment between the BIM model and the specification text to ensure the comprehensiveness and accuracy of the review results. Solving these two problems requires not only the support of advanced natural language processing technology and deep learning algorithms, but also a deep understanding and grasp of architectural design specifications.

[0005] Therefore, how to realize automatic identification and reasoning of complex regulatory logic through the automated compliance review method of building information model driven by large language model, and at the same time achieve efficient and accurate alignment between BIM model and specification text, so as to greatly improve the automation level and accuracy of BIM compliance review, has become an urgent problem to be solved. Summary of the invention

[0006] In view of the above-mentioned deficiencies in the prior art, the present invention provides a large language model-driven automated compliance review method for building information models, which can realize automatic recognition and reasoning of complex regulatory logic, and at the same time achieve efficient and accurate alignment between BIM models and specification texts, thereby greatly improving the automation level and accuracy of BIM compliance reviews.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] The method for automated compliance review of building information models driven by a large language model includes the following steps:

[0009] S1. Obtain the standard text of the building design code and the BIM model to be reviewed;

[0010] S2. Use a large language model to match the standard text with predefined standard entities of named entity recognition (NER), and use BIO annotation rules to identify the matched entities to obtain the preliminarily identified entities.

[0011] S3. Verify and optimize each initially identified entity through a preset self-correction method to obtain each verified entity;

[0012] S4. Based on a predefined IFC class tag set of the BIM model, each verified entity is mapped to a corresponding IFC class to obtain an IFC class recognition result; the IFC class tag set includes multiple candidate IFC classes of the BIM model, which are used to identify and classify different building elements or components;

[0013] S5. docking the IFC class identification result with the specification text body, wherein the docking includes matching and associating; and based on the docking result, extracting various attribute information related to the specification text from the BIM model to be reviewed;

[0014] S6, according to the structure and format of the entity corresponding to each IFC class recognition result in the standard text, convert each extracted attribute content into a Text text with a corresponding structure and format;

[0015] S7. Perform a matching check and compliance review on the Text obtained in S6 based on the specification text to obtain the compliance result of the BIM model to be reviewed.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] 1. Efficient entity recognition and preliminary verification. This method uses a large language model for named entity recognition (NER), combined with BIO annotation rules, to accurately and quickly preliminarily identify key entities from the standard text. Compared with the methods in the prior art that rely on rule bases or shallow semantic parsing, the large language model, with its powerful language understanding and generation capabilities, can more deeply understand the semantics of standard texts, thereby more accurately identifying entities. The pre-identified entities are verified and optimized through the preset self-correction method, further improving the accuracy and reliability of the recognition. This step effectively reduces subsequent review issues caused by entity recognition errors and improves the efficiency and accuracy of the entire review process.

[0018] 2. Accurate IFC class mapping. This method maps the verified entities to the corresponding IFC classes based on the predefined IFC class tag set of the BIM model. This step achieves semantic alignment between BIM model elements and specification provisions, laying the foundation for subsequent attribute information extraction and compliance review. Compared with the existing technology, this method can more accurately identify the correspondence between entities and IFC classes through the deep semantic understanding of the large language model, avoiding the deviation of review results caused by mapping errors.

[0019] 3. Efficient attribute information extraction and text conversion. Based on the IFC class recognition results, this method extracts attribute information related to the specification text from the BIM model and converts it into a text with the corresponding structure and format. This step realizes the conversion of BIM model data into the specification text format, which facilitates subsequent matching checks and compliance reviews. Compared with traditional manual extraction or rule-based methods, this method can process large amounts of data more efficiently while ensuring the accuracy and completeness of the extracted information.

[0020] 4. Automated compliance review. This method performs a matching check and compliance review on the converted Text text based on the standard text to obtain the compliance result of the BIM model to be reviewed. This step realizes the automation of BIM compliance review and greatly improves the review efficiency. Compared with the existing technology, this method can more accurately identify and process the complex logic and implicit rules in the standard text through the deep semantic understanding and reasoning ability of the large language model, thereby improving the accuracy and reliability of the review results.

[0021] In summary, this method realizes the automatic recognition and reasoning of complex regulatory logic through efficient entity recognition and preliminary verification, accurate IFC class mapping, efficient attribute information extraction and text conversion, and automated compliance review, and at the same time achieves efficient and accurate alignment between BIM models and specification texts. This technical solution not only greatly improves the automation and accuracy of BIM compliance review, but also reduces the cost and risk of manual review, providing strong support for the digital transformation and intelligent management of the construction industry.

[0022] Preferably, in S2, the predefined standard entities of named entity recognition (NER) include buildings, building components, and building spaces.

[0023] This setting enhances the pertinence and accuracy of identification, promotes the depth and breadth of semantic understanding, and improves the automation of the review process. These effects work together on the entire BIM compliance review process, providing strong support for the digital transformation and intelligent management of the construction industry.

[0024] Preferably, in S3, the preset self-correction method includes: cross-validating and optimizing the initially identified entities through another large language model combined with the thinking chain CoT technology; or, combining with an artificially preset tag set, performing comparison verification and error correction on the entities initially identified and output by the large language model.

[0025] The thinking chain technology guides the model to reason step by step, ensuring multi-level verification of the accuracy of each recognized entity. By guiding the model to reason and verify from different angles, the errors that may occur in the initial recognition of the model are corrected. If problems are found in the recognition results during the cross-validation process, the system will perform multiple rounds of iterative corrections until the entity results meet the preset standards or the maximum number of verification attempts is reached. The final result returns the optimized result in JSON format, which includes three parts: a description of the thinking process and reasoning steps for verification and optimization; an optimized entity list, which only contains the original entities that meet the standards and the final evaluation results of whether they meet the standards. In specific implementation, manual preset mark comparison can also be used as an effective error correction method. Combined with the manually preset mark set, the entities output by the model are checked and corrected. The mark set includes standard entities and their annotations that have been manually reviewed to help the model identify potential misidentification or omissions. This process combines manual marking with automated verification to ensure that the entity annotation results meet the standard requirements of the BIM model.

[0026] Such a setting, 1. By introducing another large language model for cross-validation, the advantages of different models in language understanding and reasoning can be used to comprehensively review the initially identified entities from multiple perspectives. This multi-model collaboration helps to discover information that a single model may miss or misunderstand, thereby improving the comprehensiveness of the verification. At the same time, combined with the thinking chain technology, the model can gradually reason and verify the correctness of the entity in a logical order, which increases the depth of verification and helps to reveal the complex semantic and logical relationships between entities.

[0027] 2. Improve the accuracy and efficiency of optimization. The thought chain technology can not only help the model understand the semantic meaning of entities, but also guide the model to accurately optimize the initially identified entities. By simulating the human thinking process, the model can identify and correct errors or deficiencies in the initial identification, thereby improving the accuracy of the entity. In addition, because the thought chain technology can guide the model to perform orderly reasoning, it avoids the blindness and inefficiency that may exist in traditional methods and improves the efficiency of optimization.

[0028] Preferably, in S4, when each verified entity is mapped to a corresponding IFC class, if an entity cannot be matched to a corresponding IFC class, a preset error response is returned for compliance review in S7; the content of the error response includes: the IFC class object cannot be found.

[0029] Such a setting, 1. Improves the accuracy of mapping. By setting up a mechanism to return an error response when a match cannot be made, the system can clearly indicate which entities have problems in the mapping process. This helps reviewers quickly locate and solve problems, avoiding deviations in subsequent review results caused by incorrect mapping. At the same time, this mechanism also prompts the system to be more cautious and accurate in the mapping process, because any mapping errors will be clearly recorded and fed back.

[0030] 2. Enhance the reliability of the review. In the BIM compliance review, it is crucial to accurately map entities to the corresponding IFC classes. Because IFC classes represent different elements or components in the BIM model, they are the basis for subsequent attribute information extraction and compliance judgment. By returning an error response, the system can ensure that only correctly mapped entities are used in subsequent review steps, thereby enhancing the reliability of the entire review process.

[0031] Preferably, in S5, each attribute information is extracted from the BIM model through the ifcopenshell tool.

[0032] With such a setting, ifcopenshell, as an open source IFC toolkit and geometry engine, is specifically used to process BIM files (especially files in IFC format). It provides a rich API interface, making it efficient and accurate to extract attribute information from BIM models. Through ifcopenshell, you can easily access various elements and attributes in the BIM model, including geometric information, material information, physical information, etc. of buildings, components, spaces, etc., so as to meet the comprehensive needs for attribute information in compliance reviews.

[0033] In addition, ifcopenshell supports multiple IFC versions (such as IFC2X3, IFC4, etc.) and can process different versions of BIM files, which enhances its compatibility in different projects and scenarios. At the same time, ifcopenshell's API design is flexible, allowing developers to customize the extraction logic of attribute information according to specific needs, thereby improving the flexibility and customizability of the extraction process.

[0034] Preferably, S3 also includes setting a corresponding prompt template according to the structure and format of each verified entity in the standard text; in S6, each attribute information is converted into a Text text in a corresponding format through the prompt template.

[0035] Such a setting,1. In S3, setting the corresponding prompt template according to the structure and format of the verified entity in the standard text can make the subsequent information conversion process more efficient. Through the preset template, the attribute information can be quickly and accurately converted into a Text text that meets the requirements of the standard text, reducing the possibility of manual intervention and errors.

[0036] 2. Ensure information consistency. The use of prompt templates can ensure that the converted text is consistent with the standard text in format, structure and style. This helps to avoid deviations or omissions in the information conversion process and improve the accuracy and completeness of the information. The converted text meets the requirements of the standard text, which is convenient for subsequent processing and analysis. For example, these texts can be input into the compliance review system for automatic review, or used to generate reports, conduct data mining, etc.

[0037] 3. In S6, the prompt template is used to convert each attribute information into a text of the corresponding format, realizing the automatic processing of information. This greatly improves the processing efficiency and reduces the time and cost of manual operation. Using the prompt template for information conversion can reduce errors caused by improper manual operation. The preset rules and formats of the template ensure the accuracy and consistency of the conversion process.

[0038] Preferably, in S7, the matching check includes: for a Text text of attribute information, matching each clause in the standard text one by one, and calculating the corresponding matching degree, and finding the clause with the highest matching degree; if the highest matching degree is higher than a preset threshold, the attribute information enters compliance review; otherwise, generating a review error message for the attribute information.

[0039] This setting, 1. Improves the accuracy of the review. By matching the text of the attribute information with the provisions in the specification text one by one and calculating the matching degree, it is possible to accurately determine whether the attribute information complies with the corresponding building specifications. This one-to-one matching method avoids review errors caused by omissions or misunderstandings of specification provisions and improves the accuracy of the review.

[0040] 2. Enhance the degree of automation of review. The matching check process can be automated, reducing the workload of manual review and improving review efficiency. Automated review can also quickly process a large amount of attribute information, which is suitable for compliance checks of large BIM projects.

[0041] 3. Provide clear review feedback. For attribute information whose matching degree is lower than the preset threshold, the system will output review error information, clearly indicating which attribute information does not meet the specifications. This clear feedback helps users quickly locate and correct problems, improving the practicality and guidance of the review.

[0042] Preferably, in S7, the compliance review includes: reviewing the Text of the attribute information with the clause with the highest matching degree, the review includes information completeness and information conformity, and obtaining the compliance result of the attribute information based on the review result.

[0043] Such a setting, 1. Improve the depth and comprehensiveness of the review. The compliance review not only focuses on the basic matching of attribute information and regulatory provisions, but also further evaluates the completeness and conformity of the information, thereby ensuring that the attribute information is not only relevant to the regulatory provisions, but also complete and accurate in content. This in-depth review helps to discover subtle problems that may exist in the attribute information and improve the comprehensiveness and accuracy of the review.

[0044] 2. Enhance the reliability of compliance judgment. Through the dual evaluation of information completeness and information compliance, it is possible to more reliably judge whether the property information meets the requirements of the building code. This reliable compliance judgment helps reduce compliance risks caused by lax review and improve the quality and safety of construction projects.

[0045] 3. Provide detailed review reports. During the compliance review process, the system will record the detailed steps and results of the review, including information completeness, information compliance, and final compliance results. These detailed review reports help users understand the compliance status of attribute information, as well as possible problems and improvement directions.

[0046] Preferably, after S6 and before S7, it also includes S61, randomly extracting a preset number of Text texts from the Text texts corresponding to each attribute information of each BIM model to be reviewed as verification instances, and performing a matching check on each Text text in the verification instance, and counting the proportion of the highest matching degree higher than a preset threshold. If the proportion is higher than the preset proportion, it is judged that the obtained standard text of the standard clause meets the applicability, and go to S7; otherwise, it is judged that it does not meet the adaptability, outputs the preset standard clause inapplicability information, and re-obtains the standard text of the new standard clause, and then returns to S2.

[0047] This setting evaluates the applicability of the specification text by randomly selecting verification examples and performing matching checks. The addition of this technical content has shown significant effects in improving the applicability of the specification text, enhancing the flexibility of the review process, reducing the risk of misjudgment and omission, improving user experience and satisfaction, and promoting the continuous updating and optimization of the specification. These effects together provide a strong guarantee for the compliance review of the BIM model and promote the digital transformation and intelligent development of the construction industry.

[0048] Preferably, in S7, the compliance result of the BIM model is output in JSON format; the content of the compliance result includes the review conclusion of each component.

[0049] With this setting, the compliance results of the BIM model are output in JSON format in step S7, and the review conclusions of each component are included. This technical setting has shown significant effects in data standardization, result clarity, subsequent processing and analysis, support for complex data structures, and improved review efficiency and transparency. These effects together provide strong support for the compliance management of BIM models and promote the digital transformation and intelligent development of the construction industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0051] Figure 1 A flowchart of the method is shown in FIG.

[0052] Figure 2 This is an example diagram of the processing process of the compliance result analysis of the BIM model in Example 1. DETAILED DESCRIPTION

[0053] The following is a further detailed description through specific implementation methods:

[0054] Embodiment 1

[0055] like Figure 1 , Figure 2 As shown, this embodiment discloses a large language model driven building information model automated compliance review method.

[0056] It should be noted that Figure 2 The implementation process is to encode the specific process of this method into an intelligent program module for processing. Figure 2 The named entity self-correction module, mapping module, BIM attribute information extraction module, BIM-Text generation module, and reasoning module are the functional modules of the corresponding processes in this method. Integrating the functions into the software module is a conventional means of intelligent systems and will not be repeated here.

[0057] The method comprises the following steps:

[0058] S1. Obtain the standard text of the building design code and the BIM model to be reviewed;

[0059] S2. Use a large language model to match the canonical text with predefined standard entities of named entity recognition (NER), and use BIO annotation rules to identify the matched entities to obtain the preliminarily identified entities.

[0060] In specific implementation, the predefined standard entities of named entity recognition (NER) include buildings, building components, and building spaces. In this way, the pertinence and accuracy of recognition are enhanced, the depth and breadth of semantic understanding are promoted, and the degree of automation of the review process is improved. These effects work together on the entire BIM compliance review process, providing strong support for the digital transformation and intelligent management of the construction industry.

[0061] S3. Through a preset self-correction method, each initially identified entity is verified and optimized to obtain each verified entity.

[0062] Among them, the preset self-correction method includes: cross-validation and optimization of the initially identified entities through another large language model combined with the Chain of Thought CoT technology. Chain of Thought CoT technology, namely Chain-of-Thought, referred to as CoT, can significantly improve the logical reasoning ability of the large language model by guiding it to answer questions step by step.

[0063] In the specific processing flow of self-correction, the model uses a predefined named entity recognition (NER) method to preliminarily process the specification text, identify entities in the text and label them. These preliminarily identified entities are input into another large language model, which combines the chain of thought (CoT) technology to cross-validate and optimize the entities. Through the chain of thought technology, the model will reason step by step to confirm whether each entity conforms to the standard representation in the building information model (BIM). The model will check whether each entity conforms to the standardized definition of building elements and components in BIM, and ensure that the semantics of the entity in the specification text is consistent with the mapping class of the BIM model. The model will correct the errors in the preliminary identification based on the reasoning results, delete the entities that do not meet the standards and supplement the missing entities. Through multiple rounds of reasoning and verification, the model gradually optimizes the accuracy and completeness of the entities and finally outputs the results that meet the requirements of the specification. Finally, all verified and optimized entities are output in JSON format, including the reasoning process of each entity, the optimized entity list and the evaluation results, to ensure that the results obtained meet the requirements of the building information model and the specification.

[0064] The thinking chain technology guides the model to reason step by step, ensuring multi-level verification of the accuracy of each recognized entity. By guiding the model to reason and verify from different angles, it corrects the errors that may occur in the model during initial recognition. If problems are found in the recognition results during the cross-validation process, the system will perform multiple rounds of iterative corrections until the entity results meet the preset standards or the maximum number of verification attempts is reached. The final result returns the optimized result in JSON format, which includes three parts: a description of the thinking process and reasoning steps for verification and optimization; an optimized entity list that only contains the original entities that meet the standards and the final evaluation results of whether they meet the standards.

[0065] In specific implementation, manually preset tag comparison can also be used as an effective error correction method. The preset self-correction method can also include: combining the manually preset tag set to check and correct the entities initially recognized and output by the large language model. The tag set includes standard entities and their annotations that have been manually reviewed to help the model identify potential misidentifications or omissions. This process combines manual tagging with automated verification to ensure that the entity annotation results meet the standard requirements of the BIM model.

[0066] In this way, by introducing another large language model for cross-validation, the advantages of different models in language understanding and reasoning can be used to comprehensively review the preliminarily identified entities from multiple perspectives. This multi-model collaboration helps to discover information that a single model may miss or misunderstand, thereby improving the comprehensiveness of the verification. At the same time, combined with the thinking chain technology, the model can gradually reason and verify the correctness of the entity in a logical order, which increases the depth of verification and helps to reveal the complex semantic and logical relationships between entities. In addition, it can also improve the accuracy and efficiency of optimization. Thinking chain technology can not only help the model understand the semantic meaning of the entity, but also guide the model to accurately optimize the preliminarily identified entities. By simulating the human thinking process, the model can identify and correct errors or deficiencies in the preliminary identification, thereby improving the accuracy of the entity. In addition, since the thinking chain technology can guide the model to reason in an orderly manner, it avoids the blindness and inefficiency that may exist in traditional methods and improves the efficiency of optimization.

[0067] S4. Based on a predefined IFC class tag set of the BIM model, each verified entity is mapped to a corresponding IFC class to obtain an IFC class recognition result; the IFC class tag set includes multiple candidate IFC classes of the BIM model, which are used to identify and classify different building elements or components.

[0068] In specific implementation, when each verified entity is mapped to the corresponding IFC class, if an entity cannot be matched to the corresponding IFC class, a preset error response is returned for S7 compliance review; the content of the error response includes: IFC class object cannot be found. In this way, by setting a mechanism to return an error response when a match cannot be made, the system can clearly indicate which entities have problems in the mapping process. This helps reviewers to quickly locate and solve problems, avoiding deviations in subsequent review results caused by incorrect mapping. At the same time, this mechanism also prompts the system to be more cautious and accurate in the mapping process, because any mapping errors will be clearly recorded and fed back. In addition, in BIM compliance review, it is crucial to accurately map entities to corresponding IFC classes. Because IFC classes represent different elements or components in the BIM model, they are the basis for subsequent attribute information extraction and compliance judgment. By returning an error response, the system can ensure that only correctly mapped entities will be used in subsequent review steps, thereby enhancing the reliability of the entire review process.

[0069] Through this process, entities such as buildings, components, and spaces in the specification text are aligned with component types in BIM, laying the foundation for subsequent BIM model attribute information extraction and compliance review. In this step, the high-quality verified entities are first mapped through LLM, and each entity is matched with a candidate class in the predefined IFC class tag set. For example, "factory" should be mapped to IfcBuilding, and "load-bearing wall" is mapped to IfcWall. The input data comes from the output of the first step, which includes the content of the clauses that have been marked as BIM-related entities. By comparing with the candidate IFC class tags, the system achieves semantic alignment between entities and IFC classes.

[0070] During the mapping process, LLM ensures that each entity is accurately mapped to an IFC class according to preset rules and does not exceed the candidate range. If the corresponding IFC class cannot be found during the matching process, a response of "object not found" will be returned. These errors will be eliminated in the final compliance review stage. The output results will be presented as the entity and corresponding IFC class annotation of each clause, and returned in JSON format. For example, the input clause "The fire resistance limit of the fire wall of the factory building with a fire resistance level of not less than 3h" will get the following mapping results: the factory building is mapped to IfcBuilding, and the firewall is mapped to IfcWall. This process not only ensures that the building entity in the specification text is consistent with the component type in the BIM model, but also provides structured input for data processing in subsequent steps.

[0071] S5. docking the IFC class identification result with the specification text body, wherein the docking includes matching and associating; and based on the docking result, extracting various attribute information related to the specification text from the BIM model to be reviewed.

[0072] These attribute data are usually identified by ifc-guid and contain a variety of detailed information about building components, such as material type, size, fire resistance, etc. For example, when a building component (such as a "concrete column") is identified, the system will obtain the specific properties of the component from the BIM model, such as "fire resistance" or "volume", etc. This attribute information will be matched with the previously identified building entity to provide basic data for subsequent text generation.

[0073] In the specific implementation, the ifcopenshell tool is used to extract various attribute information from the BIM model. As an open source IFC toolkit and geometry engine, ifcopenshell is specifically used to process BIM files (especially files in IFC format). It provides a rich API interface, making it efficient and accurate to extract attribute information from the BIM model. Through ifcopenshell, you can easily access various elements and attributes in the BIM model, including geometric information, material information, physical information, etc. of buildings, components, spaces, etc., so as to meet the comprehensive requirements for attribute information in compliance reviews. In addition, ifcopenshell supports multiple IFC versions (such as IFC2X3, IFC4, etc.) and can process different versions of BIM files, which enhances its compatibility in different projects and scenarios. At the same time, the API design of ifcopenshell is flexible, allowing developers to customize the extraction logic of attribute information according to specific needs, thereby improving the flexibility and customizability of the extraction process.

[0074] Those skilled in the art may also choose to familiarize themselves with other BIM data processing tools, such as Revit API or Solibri, which will not be described here.

[0075] S6. According to the structure and format of the entity corresponding to each IFC class recognition result in the standard text, the extracted attribute content is converted into a Text text with the corresponding structure and format.

[0076] In specific implementation, S3 also includes setting a corresponding prompt template according to the structure and format of each verified entity in the standard text; in S6, each attribute information is converted into a Text text in a corresponding format through the prompt template. In this way, in S3, the corresponding prompt template is set according to the structure and format of the verified entity in the standard text, which can make the subsequent information conversion process more efficient. Through the preset template, the attribute information can be quickly and accurately converted into a Text text that meets the requirements of the standard text, reducing the possibility of manual intervention and errors. It can also ensure information consistency. The use of the prompt template can ensure that the converted Text text is consistent with the standard text in format, structure and style. This helps to avoid deviations or omissions in the information during the conversion process and improve the accuracy and completeness of the information. The converted Text text meets the requirements of the standard text and is convenient for subsequent processing and analysis. For example, these texts can be input into the compliance review system for automatic review, or used to generate reports, perform data mining, etc. In S6, each attribute information is converted into a Text text in a corresponding format through the prompt template, realizing the automatic processing of information. This greatly improves processing efficiency and reduces the time and cost of manual operations. Using prompt templates to convert information can reduce errors caused by improper manual operation. The preset rules and formats of the template ensure the accuracy and consistency of the conversion process.

[0077] Through the designed prompt template, the specification clauses are combined with the extracted attribute data, and a concise and clear description related to them is generated. For example, when the clause mentions "the fire resistance limit of the columns of the factory building with a fire resistance level of three shall not be less than 2 hours", the system will splice the text data that meets the requirements of the clause based on the attribute information in the BIM model, such as the fire resistance limit. The core of this process is to ensure that the information in the specification clauses is highly aligned with the data in the BIM model and presented in a smooth manner. This step avoids unnecessary reasoning or evaluation through precise text splicing, and only focuses on the extraction and serialization of evidence. For each entity or component to be reviewed, the system will match its relevant attributes (such as fire resistance limit, size, etc.) with the requirements in the specification clauses, generate corresponding descriptions, and store them as data in JSON format for further processing. Throughout the process, the system strictly follows the content of the specification clauses, does not make compliance judgments, and simply splices the relevant entity information and outputs it in a standard text form. In this way, the system can effectively provide accurate and detailed text data support for subsequent compliance reviews while maintaining data consistency and structure.

[0078] S7. Perform a matching check and compliance review on the Text obtained in S6 based on the specification text to obtain the compliance result of the BIM model to be reviewed.

[0079] In specific implementation, the compliance results of the BIM model are output in JSON format; the content of the compliance results includes the review conclusions of each component. In this way, the compliance results of the BIM model are output in JSON format, and the review conclusions of each component are included. This technical setting has shown significant effects in data standardization, result clarity, subsequent processing and analysis, support for complex data structures, and improved review efficiency and transparency. These effects together provide strong support for the compliance management of the BIM model and promote the digital transformation and intelligent development of the construction industry.

[0080] In specific implementation, the matching check includes: for a Text text of attribute information, matching each clause in the specification text one by one, and calculating the corresponding matching degree to find the clause with the highest matching degree; if the highest matching degree is higher than the preset threshold, the attribute information enters compliance review; otherwise, a review error message of the attribute information is generated.

[0081] This can improve the accuracy of the review. By matching the Text of the attribute information with the provisions in the specification text one by one and calculating the matching degree, it is possible to accurately determine whether the attribute information complies with the corresponding building specifications. This one-to-one matching method avoids review errors caused by omissions or misunderstandings of specification provisions and improves the accuracy of the review. The degree of automation of the review can also be enhanced. The matching check process can be automated, reducing the workload of manual review and improving review efficiency. Automated review can also quickly process a large amount of attribute information, which is suitable for compliance checks of large BIM projects. Clear review feedback can also be provided. For attribute information with a matching degree lower than the preset threshold, the system will output a review error message, clearly indicating which attribute information does not meet the specifications. This clear feedback helps users quickly locate and correct problems, improving the practicality and guidance of the review.

[0082] The compliance review includes: reviewing the Text of the attribute information with the clause with the highest matching degree, the review includes information completeness and information compliance, and obtaining the compliance result of the attribute information based on the review result.

[0083] In this way, the depth and comprehensiveness of the review can be improved. The compliance review not only focuses on the basic matching of the property information with the specification provisions, but also further evaluates the information completeness and information compliance, so as to ensure that the property information is not only related to the specification provisions, but also complete and accurate. This in-depth review helps to discover subtle problems that may exist in the property information and improve the comprehensiveness and accuracy of the review. Enhance the reliability of compliance judgment. Through the dual evaluation of information completeness and information compliance, it is possible to more reliably judge whether the property information meets the requirements of the building specification. This reliable compliance judgment helps to reduce the compliance risks caused by lax review and improve the quality and safety of construction projects. In addition, detailed review reports can be provided. During the compliance review process, the system will record the detailed steps and results of the review, including information completeness, information compliance and the final compliance results. These detailed review reports help users understand the compliance status of the property information, as well as possible problems and improvement directions.

[0084] This method uses a large language model for named entity recognition (NER), combined with BIO annotation rules, to accurately and quickly identify key entities from the standard text. Compared with the methods in the prior art that rely on rule bases or shallow semantic parsing, the large language model can more deeply understand the semantics of the standard text with its powerful language understanding and generation capabilities, thereby more accurately identifying entities. The preliminarily identified entities are verified and optimized through the preset self-correction method, further improving the accuracy and reliability of recognition. This step effectively reduces the subsequent review problems caused by entity recognition errors and improves the efficiency and accuracy of the entire review process. In addition, this method maps the verified entities to the corresponding IFC classes based on the predefined IFC class tag set of the BIM model. This step achieves semantic alignment between BIM model elements and standard texts, laying the foundation for subsequent attribute information extraction and compliance review. Compared with the prior art, this method can more accurately identify the correspondence between entities and IFC classes through the deep semantic understanding of the large language model, avoiding the deviation of review results caused by mapping errors.

[0085] According to the IFC class recognition results, this method extracts attribute information related to the specification text from the BIM model and converts it into a Text text with the corresponding structure and format. This step realizes the conversion of BIM model data into the specification text format, which is convenient for subsequent matching checks and compliance reviews. Compared with traditional manual extraction or rule-based methods, this method can process large amounts of data more efficiently while ensuring the accuracy and completeness of the extracted information. In addition, this method performs matching checks and compliance reviews on the converted Text text based on the specification text to obtain the compliance results of the BIM model to be reviewed. This step realizes the automation of BIM compliance review and greatly improves the review efficiency. Compared with the existing technology, this method can more accurately identify and process complex logic and implicit rules in the specification provisions through the deep semantic understanding and reasoning capabilities of the large language model, thereby improving the accuracy and reliability of the review results.

[0086] This method realizes the automatic recognition and reasoning of complex regulatory logic through efficient entity recognition and preliminary verification, accurate IFC class mapping, efficient attribute information extraction and text conversion, and automated compliance review, while achieving efficient and accurate alignment between BIM models and regulatory texts. This technical solution not only greatly improves the automation and accuracy of BIM compliance review, but also reduces the cost and risk of manual review, providing strong support for the digital transformation and intelligent management of the construction industry.

[0087] Embodiment 2

[0088] Different from the first embodiment, this embodiment further includes S61 after S6 and before S7, randomly extracting a preset number of Text texts as verification instances from the Text texts corresponding to each attribute information of each BIM model to be reviewed, and performing a matching check on each Text text in the verification instance, and counting the proportion of the highest matching degree higher than a preset threshold. If the proportion is higher than the preset proportion, it is judged that the obtained standard text of the standard clause meets the applicability, and go to S7; otherwise, it is judged that it does not meet the adaptability, outputs the preset standard clause inapplicability information, and re-obtains the standard text of the new standard clause, and then returns to S2.

[0089] In this way, by randomly selecting verification examples and performing matching checks to evaluate the applicability of the specification text, the addition of this technical content has shown significant effects in improving the applicability of the specification text, enhancing the flexibility of the review process, reducing the risk of misjudgment and omission, improving user experience and satisfaction, and promoting the continuous updating and optimization of the specification. These effects together provide a strong guarantee for the compliance review of the BIM model and promote the digital transformation and intelligent development of the construction industry.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.

Claims

1. A large language model driven building information model automated compliance review method, characterized in that: The following steps are involved: S1. Obtain the standard text of the building design code and the BIM model to be reviewed; S2. Use a large language model to match the standard text with predefined standard entities of named entity recognition (NER), and use BIO annotation rules to identify the matched entities to obtain the preliminarily identified entities. S3. Verify and optimize each initially identified entity through a preset self-correction method to obtain each verified entity; S4. Based on a predefined IFC class tag set of the BIM model, each verified entity is mapped to a corresponding IFC class to obtain an IFC class recognition result; the IFC class tag set includes multiple candidate IFC classes of the BIM model, which are used to identify and classify different building elements or components; S5. docking the IFC class identification result with the specification text body, wherein the docking includes matching and associating; and based on the docking result, extracting various attribute information related to the specification text from the BIM model to be reviewed; S6, according to the structure and format of the entity corresponding to each IFC class recognition result in the standard text, convert each extracted attribute content into a Text text with a corresponding structure and format; S7. Perform a matching check and compliance review on the Text obtained in S6 based on the specification text to obtain the compliance result of the BIM model to be reviewed.

2. The method for automated compliance review of building information models driven by a large language model as claimed in claim 1, characterized in that: In S2, the predefined standard entities of named entity recognition (NER) include buildings, building components, and building spaces.

3. The large language model driven automated compliance review method for building information models according to claim 1, characterized in that: In S3, the preset self-correction method includes: cross-validating and optimizing the initially identified entities through another large language model combined with the thinking chain CoT technology; or, combining with a manually preset tag set, comparing and correcting the entities initially identified and output by the large language model.

4. The large language model driven automated compliance review method for building information models according to claim 1, characterized in that: In S4, when each verified entity is mapped to the corresponding IFC class, if an entity cannot be matched to the corresponding IFC class, a preset error response is returned for compliance review in S7; The content of the error response includes: the IFC class object cannot be found.

5. The large language model driven automated compliance review method for building information models according to claim 1, characterized in that: In S5, the ifcopenshell tool is used to extract various attribute information from the BIM model.

6. The large language model driven automated compliance review method for building information models according to claim 1, characterized in that: S3 also includes setting a corresponding prompt template according to the structure and format of each verified entity in the standard text; In S6, each attribute information is converted into a Text text of a corresponding format through the prompt template.

7. The large language model driven automated compliance review method for building information models according to claim 6, characterized in that: In S7, the matching check includes: for a Text text of attribute information, matching each clause in the specification text one by one, and calculating the corresponding matching degree to find the clause with the highest matching degree; if the highest matching degree is higher than the preset threshold, the attribute information enters compliance review; otherwise, generating a review error message for the attribute information.

8. The large language model driven automated compliance review method for building information models according to claim 7, characterized in that: In S7, the compliance review includes: reviewing the Text of the attribute information with the clause with the highest matching degree, the review includes information completeness and information compliance, and obtaining the compliance result of the attribute information based on the review result.

9. The large language model driven automated compliance review method for building information models according to claim 7, characterized in that: After S6 and before S7, it also includes S61, randomly extracting a preset number of Text texts from the Text texts corresponding to each attribute information of each BIM model to be reviewed as verification instances, and performing matching checks on each Text text in the verification instance respectively, and counting the proportion of the highest matching degree higher than a preset threshold value, if the proportion is higher than the preset proportion, it is determined that the obtained standard text of the standard clause meets the applicability, and go to S7; Otherwise, it is judged that the adaptability is not met, and the preset standard clause is not applicable. The standard text of the standard clause of the letter is re-obtained, and then return to S2.

10. The large language model driven automated compliance review method for building information models according to claim 7, characterized in that: In S7, the compliance result of the BIM model is output in JSON format; the content of the compliance result includes the review conclusion of each component.

Images