Hybrid reasoning BIM (Building Information Modeling) design achievement auditing method for complex standard specification provisions

Abstract

The invention discloses a hybrid reasoning BIM (Building Information Modeling) design achievement auditing method for complex standard specification provisions. The method comprises the following steps: S1, defining a grammar structure and an entity class in a standard specification ontology; s2, constructing a standard specification knowledge graph based on the natural language large model; s3, constructing a mapping table; s4, constructing a design result file and text description information; s5, generating an executable rule statement; and S6, performing hybrid reasoning based on a rule reasoning engine and a natural language large model cue word technology. The method comprises the following steps: disassembling a standard text by using a natural language large model cue word technology, and constructing a knowledge graph; the natural language standard specification text is converted into an executable rule, and compliance auditing of the BIM design result is supported; and a rule inference engine and a large language model are combined to perform hybrid inference, so that the BIM model auditing precision is improved.

Description

Technical Field

[0001] This invention relates to the field of railway engineering management technology, specifically to a hybrid reasoning BIM design outcome review method for complex standard and specification clauses. Background Technology

[0002] Compliance review of Building Information Modeling (BIM) is a core aspect of ensuring the quality of engineering design and construction, especially in complex infrastructure sectors such as railways, where strict adherence to multi-level, dynamically updated standards and specifications is required. Traditional review methods primarily rely on manual parsing of specification texts and encoding them into executable rules, which are then matched with the BIM model using a rule engine. However, standard specification texts are highly specialized and semantically complex, often containing intricate semantic relationships. Developing static rules not only requires significant human and material resources but also suffers from the risk of overlooking key entities and relationships during manual analysis, resulting in a rule base that fails to cover the actual standard specification provisions. Even with intelligent natural language standard specification conversion methods, the limitations of existing AI techniques and the complexity of natural language standards mean that the generated executable rules may not necessarily conform to the original intent of the standard specification provisions.

[0003] In recent years, although Natural Language Modeling (NLP) technology has made breakthroughs in text understanding and generation, existing technologies have failed to achieve a closed-loop transformation from unstructured, standardized text to executable review rules. The informal results output by NLP models are difficult to directly drive inference engines. Furthermore, the current foundation of NLP models relies on probabilistic generation-based deep neural network methods, which struggle to guarantee the accuracy of compliance review. Executable rule conversion methods based on NLP models face challenges in conversion accuracy and understanding complex semantics. Therefore, it is necessary to design multiple inference mechanisms to minimize the error rate of compliance review. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a hybrid reasoning BIM design outcome review method for complex standard and specification clauses, in order to reduce the error rate of compliance review.

[0005] Therefore, the present invention adopts the following technical solution:

[0006] A hybrid reasoning-based BIM design deliverable review method for complex standard and specification clauses includes the following steps:

[0007] S1, Define the syntax structure and entity classes in the standard specification ontology. The syntax structure and entity classes constitute the standard specification ontology. The syntax structure is a structured statement consisting of a series of nested "IF-THEN" statements.

[0008] S2, a standard and specification knowledge graph built on a large natural language model;

[0009] S3. Semantically align the content in the standard specification knowledge graph formed in S2 with the corresponding content in Jena rule statements and manually reviewed BIM design results, and construct a mapping table between the component names and constraints in the standard specification clauses and the corresponding component names and Jena operator names in the manually reviewed BIM design results.

[0010] S4 iterates through the standard specification knowledge graph constructed in S2, refers to the mapping table constructed in S3, and obtains the BIM component names and attributes involved in each branch of the knowledge graph. On the one hand, it calls the rdflib library to construct the OWL format BIM design result file, and on the other hand, it uses the BIM model reading tool to construct and read the design result information described by JSON text.

[0011] S5: Traverse each branch of the standard specification knowledge graph obtained in S2, refer to the mapping table constructed in S3, and generate executable rule statements;

[0012] S6, based on the rule-based reasoning engine, performs reasoning review on the executable rule-based reasoning statements generated by S5 and the BIM design deliverables generated by S4; it also uses natural language large model prompts to review the design deliverables information generated by S4 and the original natural language form of the specification clauses; and finally, it performs manual review on the reasoning results of both to obtain the final BIM design deliverable review result.

[0013] In step S1 above, entity classes are constructed for the components, attributes, and constraints in the standard specification clauses, forming the specific components of the "IF-THEN" syntax structure.

[0014] The specific steps of S2 are as follows:

[0015] S21, using the prompting word technology of the natural language large model to complete and clearly display the "or" and "and" relationships omitted or represented by symbols in the standard specification, and placing the concluding sentences in the text at the end of the rewritten statement;

[0016] S22, using natural language large model prompting word technology, further processes the standard specification clauses completed in S21, and breaks down the completed text into independent and complete clauses according to the "or" relationship;

[0017] S23 utilizes natural language large model prompting word technology to transform each clause obtained in S22 into a structured triple, ensuring that each triple consists of one or more triples that satisfy the "and" relation.

[0018] S24 involves manually reviewing the "or" and "and" relationships of a series of triples obtained from S22 to ensure that the content and order of the triples are correct and conform to the original intent of the standard specification. The triples form a nested "IF-THEN" syntax structure, which constitutes part of the standard specification knowledge graph. All clauses are processed and identical IfStatement nodes are merged to finally form a complete standard specification knowledge graph.

[0019] The natural language model prompts used in step S21 above to complete the standard specification clauses are:

[0020] -------------

[0021] Imagine you are a semantic analysis expert for standard specification clauses. Now define an AND relation: all conditions must be satisfied simultaneously. Define an OR relation: the conditions are sequential, and satisfying any one of them is sufficient. You need to find the omitted AND and OR relations in the text and rewrite them into clauses connected by either "OR" or "AND".

[0022] # Require:

[0023] ##1. If the standard or specification clause contains clauses, please determine the "or" or "and" relationship that should be used between the clauses based on the content of the clause, and connect all the clauses into a whole using "or" or "and";

[0024] ##2. Please analyze the concluding sentences in the standard and specification clauses, place them at the end of the rewritten sentence, and describe them in the form of "Conclusion: + concluding sentence";

[0025] ##3. Punctuation marks indicating parallelism in the text should also be analyzed to determine whether they function as "or" or "and". If they do, they should be replaced with "or" or "and". The parts separated by punctuation should be supplemented with corresponding verbs to ensure grammatical correctness.

[0026] Given the sentence "{input}", output only the modified sentence.

[0027] -------------”;

[0028] When using the above prompts to complete a clause, simply replace "{input}" in the prompt with the standard specification clause to be completed, and send the entire prompt to the natural language processing model to obtain the completed standard specification clause.

[0029] The natural language large model prompt words used in step S22 above to decompose the completed normative clauses are:

[0030] -------------

[0031] Suppose you are a semantic analysis expert for standard specification clauses. Now define the AND relation: all conditions must be satisfied simultaneously. Define the OR relation: conditions are sequential; satisfying any one condition is sufficient. Analyze the semantics of a given standard specification clause, breaking it down into multiple independent clauses according to the OR relations.

[0032] # Require:

[0033] ##1. Ensure that all constraints within each clause are connected by an OR relationship;

[0034] ##2. Ensure that each clause can lead to the conclusion of the clause completely and independently.

[0035] The given text is: "{input}";

[0036] -------------”;

[0037] When using the above prompts to split clauses, replace {input} with the clause to be split to obtain the target clause.

[0038] The natural language large model prompts used in step S23 above to convert clauses into triples are as follows:

[0039] -------------

[0040] Suppose you have an entity-relation triple extraction model. First, determine what entities (subject and object) and relation types (relation) exist in the text. Then, extract the subject and object from the sentence and form triples in the form (subject, relation, object).

[0041] Define the AND relation: all conditions must be satisfied simultaneously. These conditions are broken down into individual triples, but together they represent a single fact. When encountering an AND relation: continue breaking down each condition, generating multiple triples, indicating that all conditions must be satisfied simultaneously.

[0042] Definition of an OR relation: Conditions are mutually ordered; satisfying any one of them is sufficient. These conditions are broken down into different branches, each representing a possible outcome. When encountering an OR relation: branching is performed, with each branch generating a set of triples representing different possibilities.

[0043] Given the sentence "{input}", output only the final triplet.

[0044] -------------”;

[0045] When using the above-mentioned natural language model prompts to transform standard specification clauses into triples, replace "{input}" with the clause to be transformed to obtain the final triples.

[0046] In step S6 above, the Jena inference engine is used for rule-based reasoning review, and the GenericRuleReasoner class of the Apache Jena library is used to perform reasoning review on the OWL format BIM design deliverables.

[0047] The natural language large model prompts used in step S6 above to review the design outcome information and the original natural language form of the specification clauses are:

[0048] -------------

[0049] You are an expert in railway BIM review. Please review the BIM design deliverables I have provided based on the standard specifications I have given. If the deliverables meet the requirements of the standard specifications, output "Qualified". If they do not meet the requirements, point out each non-compliant item and provide the reason for each non-compliance. The standard specification provided is "{rule}", and the BIM design deliverables are "{info}". What is your review result?

[0050] -------------”;

[0051] When using the above prompts to review the compliance of BIM design deliverables, replace "{rule}" with the original natural language form of the specification clauses, replace "{info}" with the design deliverable information described in JSON text generated by S4, input the natural language large model, and obtain the reasoning results.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] 1. This invention is based on ontology technology, that is, based on the semantic characteristics of standard specification texts, to construct an ontology for the formal transformation of standard specifications, providing data structure support for the subsequent decomposition and formal transformation of natural language standard specification clauses.

[0054] 2. This invention is based on natural language large model prompt word technology, which decomposes standard specification text in natural language form, constructs a standard specification knowledge graph, and uses the standard specification knowledge graph to realize the formal transformation from natural language standard specification text to executable rule statements. This facilitates the computer to use the transformed executable rule statements to conduct compliance review of BIM design results in IFC form that have been manually screened through a rule reasoning engine.

[0055] 3. This invention utilizes a rule-based reasoning engine and a large language model to overcome the problems of inaccurate conversion and difficulty in understanding complex semantics in the process of converting natural language rule texts into executable rule statements through hybrid reasoning, thereby improving the accuracy of compliance review of BIM model design results. Attached Figure Description

[0056] Figure 1 This is a flowchart of the review method of the present invention;

[0057] Figure 2 This is a general framework diagram of the review method of the present invention;

[0058] Figure 3 This is a schematic diagram of the standard specification knowledge graph ontology constructed in an embodiment of the present invention. Detailed Implementation

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the following embodiments are only some embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Example

[0061] like Figure 1 As shown, the hybrid reasoning BIM design outcome review method for complex standard and specification clauses of the present invention includes the following steps:

[0062] S1, Construct the standard specification ontology, such as Figure 2 As shown, it includes the following steps:

[0063] S11, Define the syntax structure in the standard specification ontology: Because standard specification clauses are used to stipulate the conditions that design results need to meet, their semantic characteristics can be abstracted into a series of nested "IF-THEN" structured statements.

[0064] S12, Define entity classes in the standard specification ontology: To describe the standard specification clauses in more detail and accurately, it is necessary to construct entity classes for the components, attributes, constraints, etc., in the standard specification clauses, forming the specific components of the "IF-THEN" syntax structure. The entity classes defined in this embodiment are shown in Table 1:

[0065] Table 1 Standard Specification Entity Class Definitions

[0066] Class Name effect Component This represents specific BIM components, such as caps, piers, beams, etc. Property It indicates the properties of a component, such as material, strength, and dimensions. Constraint This indicates requirements or limitations on components or properties, such as not exceeding 1.1 meters or not being equal to "indoor enclosed stairwell". Condition Relational entity classes establish associations between components or attributes and constraints. Action Like Condition, it is a relational entity class that establishes an association between components or attributes and constraints. However, it is only used at the end of a nested decision branch to indicate that the branch has ended.

[0067] The standard specification ontology is composed of the syntax structure in the standard specification ontology defined in S11 and the entity classes in the standard specification ontology defined in S12.

[0068] Let IfStatement represent a single "IF-THEN" syntax structure. To represent nested relationships, the following syntax is defined:

[0069] (1) An lfStatement contains a Condition, and a Condition consists of a Property or Component and a Constraint or Action.

[0070] (2) When Condition is the first lfStatement, Condition consists of a Component and a Constraint; when Condition is the last lfStatement, Condition consists of a Property and an Action. If Condition contains an Action, it means that the decision branch ends here.

[0071] (3) An lfStatement can contain multiple lfStatements, representing branches.

[0072] S2, based on a large natural language model, constructs a standard and standardized knowledge graph, such as... Figure 2 As shown, it includes the following steps:

[0073] S21 utilizes prompt word technology from a large natural language model to complete standard and specification clauses.

[0074] Using the prompting technology of natural language processing models, the "or" and "and" relationships omitted or represented by symbols (such as commas) in standard and specification clauses are completed and explicitly displayed. Furthermore, concluding sentences in the clauses (such as "When the following conditions are met, the area of ​​the fire compartment shall not exceed 10,000 m²") are also displayed. 2The phrase "1. xxxxx; 2. xxxxxx" indicates that the fire compartment area should not exceed 10,000 m². 2 The ) will be placed at the end of the rewritten statement.

[0075] The natural language large-scale model prompts used in this embodiment to complete standard specification clauses are as follows:

[0076] -------------

[0077] Imagine you are a semantic analysis expert for standard specification clauses. Now define an AND relation: all conditions must be satisfied simultaneously. Define an OR relation: the conditions are sequential; satisfying any one of them is sufficient. You need to find the omitted OR and AND relations in the text and rewrite them into clauses connected by either OR or AND.

[0078] # Require:

[0079] ##1. If the standard or specification clause contains clauses, please determine the "or" or "and" relationship that should be used between the clauses based on the content of the clause, and connect all the clauses into a whole using "or" or "and";

[0080] ##2. Please analyze the concluding sentences in the standard and specification clauses, place them at the end of the rewritten sentence, and describe them in the form of "Conclusion: + concluding sentence";

[0081] ##3. Punctuation marks indicating parallelism in the text should also be analyzed to determine whether they function as "or" or "and". If they do, they should be replaced with "or" or "and". The parts separated by punctuation should be supplemented with corresponding verbs to ensure grammatical correctness.

[0082] Given the sentence "{input}", output only the modified sentence.

[0083] -------------

[0084] When using the above prompts to complete a clause, simply replace "{input}" in the prompt with the standard specification clause to be completed, and send the entire prompt to the natural language processing model to obtain the completed standard specification clause.

[0085] For example, when "{input}" means "The building area of ​​each fire compartment in the waiting area and concourse of a railway passenger station shall not exceed 10,000 m² when the following conditions are met:",

[0086] 1. Located on the ground floor, a single-story elevated floor, or a second floor with half of its direct external evacuation exits and an enclosed indoor stairwell.

[0087] 2. The system is equipped with an automatic sprinkler system, smoke extraction facilities, and an automatic fire alarm system. Using the aforementioned natural language processing model prompts, the input clause can be translated into: "The waiting area and concourse of a railway passenger station are located on the ground floor or a single-story elevated floor, or on the second floor with half of the direct external evacuation exits and enclosed indoor stairwells, and are equipped with an automatic sprinkler system, smoke extraction facilities, and an automatic fire alarm system. Therefore, the building area of ​​each fire compartment should not exceed 10,000 m²."

[0088] S22 utilizes natural language large model prompting word technology to further process the standard specification clauses completed in S21, breaking down the completed text into independent and complete clauses according to the "or" relationship.

[0089] The natural language large-scale model prompts used in this embodiment to decompose the completed normative clauses are as follows:

[0090] -------------

[0091] Suppose you are a semantic analysis expert for standard specification clauses. Now define the AND relation: all conditions must be satisfied simultaneously. Define the OR relation: conditions are sequential; satisfying any one condition is sufficient. Analyze the semantics of a given standard specification clause, breaking it down into multiple independent clauses according to the OR relations.

[0092] # Require:

[0093] ##1. Ensure that all constraints within each clause are connected by an OR relationship;

[0094] ##2. Ensure that each clause can lead to the conclusion of the clause completely and independently.

[0095] The given text is: "{input}";

[0096] -------------

[0097] When using the above prompts to split clauses, replacing {input} with the clause to be split will yield the target clause.

[0098] For example, when “{input}” is replaced with “The waiting area and assembly hall of a railway passenger station are located on the first floor or a single-story elevated floor, or on the second floor with half of the direct external evacuation exits and enclosed indoor stairwells, and equipped with an automatic sprinkler system, smoke exhaust facilities, and an automatic fire alarm system, it is concluded that: the building area of ​​each fire compartment should not exceed 10,000 m².”, the prompt words in the above natural language model can be broken down into three clauses, as follows:

[0099] Clause 1: The waiting area and passenger hall of the railway passenger station are located on the ground floor and are equipped with an automatic sprinkler system, smoke exhaust facilities and an automatic fire alarm system. Therefore, the building area of ​​each fire compartment should not exceed 10,000 m².

[0100] Clause 2: The waiting area and passenger hall of the railway passenger station are located on a single-story elevated floor and are equipped with an automatic sprinkler system, smoke exhaust facilities and an automatic fire alarm system. Therefore, the building area of ​​each fire compartment should not exceed 10,000 m².

[0101] Clause 3: The waiting area and assembly hall of the railway passenger station are equipped with half of the direct external evacuation exits and two floors with enclosed indoor stairwells. They are also equipped with automatic sprinkler systems, smoke exhaust facilities and automatic fire alarm systems. Therefore, the building area of ​​each fire compartment should not exceed 10,000 m².

[0102] S23 utilizes natural language large model prompting technology to transform each clause obtained in S22 into a structured triple, ensuring that each triple consists of one or more triples that satisfy the "AND" relation.

[0103] The natural language large model prompts used in this embodiment to convert clauses into triples are as follows:

[0104] -------------

[0105] Suppose you have an entity-relation triple extraction model. First, determine what entities (subject and object) and relation types (relation) exist in the text. Then, extract the subject and object from the sentence and form triples in the form (subject, relation, object).

[0106] Define the AND relation: all conditions must be satisfied simultaneously. These conditions are broken down into individual triples, but together they represent a single fact. When encountering an AND relation: continue breaking down each condition, generating multiple triples, indicating that all conditions must be satisfied simultaneously.

[0107] Definition of an OR relation: Conditions are mutually ordered; satisfying any one of them is sufficient. These conditions are broken down into different branches, each representing a possible outcome. When encountering an OR relation: branching is performed, with each branch generating a set of triples representing different possibilities.

[0108] Given the sentence "{input}", output only the final triplet.

[0109] -------------

[0110] When using the above-mentioned natural language model prompts to convert standard specification clauses into triples, replacing "{input}" with the clause to be converted will yield the final triples.

[0111] For example, when the clause to be transformed is "The waiting area and assembly hall of a railway passenger station are located on the ground floor and are equipped with an automatic sprinkler system, smoke exhaust facilities, and an automatic fire alarm system, it is concluded that the building area of ​​each fire compartment should not exceed 10,000 m²," the set of triplets that can be obtained through the above natural language model prompts are: "1. (The waiting area and assembly hall of a railway passenger station are located on the ground floor); 2. (The waiting area and assembly hall of a railway passenger station are equipped with an automatic sprinkler system); 3. (The waiting area and assembly hall of a railway passenger station are equipped with smoke exhaust facilities); 4. (The waiting area and assembly hall of a railway passenger station are equipped with an automatic fire alarm system); 5. (The waiting area and assembly hall of a railway passenger station, the building area of ​​the fire compartment, should not exceed 10,000 m²)."

[0112] S24 involves manually reviewing the "OR" and "AND" relationships of the series of triples obtained in S22 to ensure that the content and order of the triples are correct and conform to the original intent of the standard specification. Each triple obtained through manual verification can be regarded as an "IF-THEN" syntactic structure, which is used to combine into part of the standard specification graph, ultimately forming the standard specification knowledge graph.

[0113] For example, the clause "The waiting area and assembly hall of a railway passenger station are located on the ground floor and are equipped with an automatic sprinkler system, smoke exhaust facilities, and an automatic fire alarm system, therefore: the building area of ​​each fire compartment should not exceed 10,000 m²." After manual review, the final set of triplets is: "1. (Railway passenger station, equipped with, waiting area and assembly hall); 2. (Waiting area and assembly hall, located on, ground floor); 3. (Waiting area and assembly hall, equipped with, automatic sprinkler system); 4. (Waiting area and assembly hall, equipped with, smoke exhaust facilities); 5. (Waiting area and assembly hall, equipped with, automatic fire alarm system); 6. (Waiting area and assembly hall, fire compartment building area, not exceeding 10,000 m²)". The nested "IF-THEN" syntax structure formed by the above set of triplets is:

[0114] "IfStatement0:{ Condition:{Component:Railway passenger station, Constraint:Waiting area and assembly hall are provided}, SubIfStatement:[IfStatement1:{ Condition:{Property:Waiting area and assembly hall, Constraint:Located on the ground floor}, SubIfStatement:[IfStatement5:{ Condition:{Property:Waiting area and assembly hall, Constraint:Automatic sprinkler system provided}, SubIfStatement:[IfStatement6:{ Condition:{Property:Waiting area and assembly hall, Constraint:Smoke exhaust facilities provided}, SubIfStatement:[IfStatement7:{ Condition:{Property:Waiting area and assembly hall, Constraint:Automatic fire alarm system provided}, SubIfStatement:[IfStatement8:{ Condition:{Property:Fire compartment building area, Action:Not greater than 10000m²}]]]]]}".

[0115] As seen in the examples above, the triples form a nested "IF-THEN" grammatical structure, constituting part of the standard specification knowledge graph. Processing all clauses and merging identical IfStatement nodes ultimately yields a complete standard specification knowledge graph, such as... Figure 3 As shown.

[0116] S3, construct the mapping table.

[0117] In this embodiment, the content in the standard specification knowledge graph formed in S2 is semantically aligned with the corresponding content in the Jena rule statements and the manually reviewed BIM design results, thus constructing a mapping table between the component names and constraints in the standard specification clauses and the corresponding component names and Jena operator names in the manually reviewed BIM design results. An example of the mapping table is shown in Table 2.

[0118] Table 2. Examples of Mapping Representations

[0119] Name in knowledge graph The name in Jena or BIM design deliverables Data format railing IfcRailing IFC Top hat TopCap IFC Fire compartment IfcSpace IFC high height IFC Not greater than le Jena operator It has has Jena operator Settings setIn Jena operator use has Jena operator

[0120] S4, construct the design deliverables and text description information.

[0121] Traverse the standard specification knowledge graph obtained in S2, refer to the mapping table constructed in S3, and obtain the BIM component names and attributes involved in each branch of the knowledge graph. On the one hand, call the rdflib library to construct the OWL format BIM design result file, and on the other hand, construct and read the design result information described by JSON text through a BIM model reading tool (such as IfcOpenShell used by IFC).

[0122] Example of design deliverable information described in the constructed JSON text: {Building Name: "Railway Passenger Station Building xxx", Review Object Attributes: {Name: "Fire Compartment Five", Location: "First Floor", Smoke Exhaust Facilities: "Yes", Automatic Sprinkler System: "Yes", Fire Alarm System: "Yes", Fire Compartment Building Area: "8000m²"}}.

[0123] S5 generates executable rule statements.

[0124] By traversing each branch of the standard specification knowledge graph obtained in S2 and referring to the mapping table constructed in S3, executable rule statements are automatically generated. In this embodiment, an example of the generated executable rule statement (Jena statement) is as follows:

[0125] [Railway Passenger Station Waiting Area and Assembly Hall Rule 1: (?a rdf:type :Railway Passenger Station Waiting Area and Assembly Hall)(?a :Fire Compartment?b)(?b :setIn "First Floor")(?b :has "Automatic Sprinkler System")(?b :has "Smoke Exhaust Facilities")(?b :has "Automatic Fire Alarm System")(?b :Building Area?c)le(?c, "10000m²")->(?a :Inference Result: Qualified)]

[0126] In the above jena inference statement, (?a rdf:type :railway passenger station waiting area and concourse) indicates that a is "railway passenger station waiting area and concourse"; (?a :fire compartment?b) indicates that b is a fire compartment of a; (?b :setIn "ground floor") indicates that the fire compartment is set on "ground floor"; (?b :has "automatic sprinkler system") indicates that b has "automatic sprinkler system"; (?b :building area?c) indicates that the building area attribute c of b is obtained; le(?c, "10000m²") means that the building area c of the fire compartment is not greater than 10000m²; (?a :inference result:qualified) means that the inference result attribute of a is assigned the value "qualified".

[0127] S6, perform mixed reasoning.

[0128] Based on the rule-based reasoning engine, the executable rule-based reasoning statements generated by S5 and the BIM design deliverables generated by S4 are subjected to reasoning review. Based on the natural language big data model, the design deliverables described in JSON text generated by S4 and the original natural language form of the specification clauses are reviewed using natural language big data model prompt words. The reasoning results of both are manually reviewed to obtain the final BIM design deliverable review result.

[0129] In this embodiment, as Figure 2 As shown, rule-based reasoning and review can be performed using the Jena reasoning engine. The GenericRuleReasoner class in the Apache Jena library can be used to perform reasoning and review on OWL format BIM design deliverables.

[0130] (The GenericRuleReasoner class is a concrete implementation of the Jena inference engine, which is a core submodule of the Apache Jena framework.)

[0131] The natural language processing model prompts used in this embodiment to review design deliverables information described in standard specifications and JSON text are as follows:

[0132] -------------

[0133] You are an expert in railway BIM review. Please review the BIM design deliverables I have provided based on the standard specifications I have given. If the deliverables meet the requirements of the standard specifications, output "Qualified". If they do not meet the requirements, point out each non-compliant item and provide the reason for each non-compliance. The standard specification provided is "{rule}", and the BIM design deliverables are "{info}". What is your review result?

[0134] -------------

[0135] When using the above prompts to review the compliance of BIM design deliverables, replace "{rule}" with the original natural language form of the specification clauses, and replace "{info}" with the design deliverable information described in JSON text generated by S4. Inputting the natural language large model will yield the inference results.

[0136] For example, when the normative clause in natural language form is "The building area of ​​each fire compartment in the waiting area and assembly hall of a railway passenger station shall not exceed 10,000 m² when it meets the following conditions: 1. Located on the ground floor, a single-story elevated floor, or a second floor with half of its direct external evacuation exits and an enclosed indoor stairwell. 2. Equipped with an automatic sprinkler system, smoke exhaust facilities, and an automatic fire alarm system."; and the design deliverable information is "{Building Name: "Railway Passenger Station Building xxx", Review Object Attributes: {Name: "Fire Compartment Five", Location: "Ground Floor", Smoke Exhaust Facilities: "None", Automatic Sprinkler System: "Yes", Fire Alarm System: "Yes", Fire Compartment Area: "8000 m²"}}", the inference result of the natural language big data model is:

[0137] Audit result: Unqualified;

[0138] Non-compliant content and reasons:

[0139] Lack of smoke extraction facilities:

[0140] • Standard specifications require that fire compartments be equipped with smoke extraction facilities.

[0141] • Design outcome information: Fire compartment five has "none" smoke exhaust facilities.

[0142] • Reason: The requirements for smoke exhaust facilities in the standard specifications are not met...."

Claims

1. A hybrid reasoning BIM design achievement auditing method for complex standard specification clauses, characterized in that, Comprise the following steps: S1, define the syntax structure and entity class in the standard specification ontology, which constitutes the standard specification ontology, and the syntax structure is a structured statement formed by a series of "IF-THEN" nesting; S2, based on natural language large model, construct standard specification knowledge graph; S3, semantic alignment of the content in the standard specification knowledge graph formed in S2 and the corresponding content in the Jena rule statement and the BIM design results audited by artificial, construct the mapping table between the component name, constraint condition in the standard specification article and the corresponding component name in the BIM design results audited by artificial and Jena operator name; S4, traverse the standard specification knowledge graph constructed in S2, refer to the mapping table constructed in S3, obtain the BIM component name and attribute involved in each branch of the knowledge graph, on the one hand, call the rdflib library to construct the BIM design results file in OWL format, on the other hand, construct and read the design results information described by Json text through BIM model reading tool; S5, traverse each branch of the standard specification knowledge graph obtained in S2, refer to the mapping table constructed in S3, and generate executable rule statement; S6, based on rule reasoning engine, reasoning auditing of the executable rule reasoning statement generated in S5 and the BIM design results file generated in S4; Using natural language large model prompt word technology, auditing the design results information generated in S4 and the original natural language form of specification article; The reasoning results of the two are audited by artificial, and the final BIM design results auditing result is obtained.

2. The hybrid-reasoning BIM design-outcome auditing method of claim 1, wherein: In S1, the components, attributes and constraints in the standard specification article are constructed as entity classes, which are the specific components of the "IF-THEN" syntax structure. 3.The hybrid inference BIM design outcome auditing method of claim 1, wherein, The specific steps of S2 are as follows: S21, use the prompt word technology of natural language large model to complete and clearly show the "or" and "and" relationship omitted or represented by symbols in the standard specification article, and the concluding sentence in the article is placed at the end of the rewritten sentence; S22, use the natural language large model prompt word technology to further process the standard specification article completed in S21, and split the supplemented text into independent and complete clauses according to the "or" relationship; S23, use the natural language large model prompt word technology to convert each clause obtained in S22 into a structured triple, which ensures that the triple is composed of one or more triples that satisfy the "and" relationship; S24, manually audit the "or" and "and" relationship of a series of triples obtained in S22 to ensure that the triple content and order are correct and consistent with the original meaning of the standard specification; Each triple forms a nested "IF-THEN" syntax structure, which constitutes part of the standard specification knowledge graph; Process all clauses and merge the same IfStatement nodes to finally form a complete standard specification knowledge graph.

4. The hybrid inference BIM design outcome auditing method of claim 3, wherein, The natural language large model prompt word used in S21 to complete the standard specification article is: "------------- Assume you are a standard specification clause semantic analysis expert; now define the AND relationship: all conditions must be met simultaneously; define the OR relationship: conditions are mutually exclusive, and meeting one is sufficient; You need to find the missing AND and OR relationships in the text and rewrite them as clauses connected by "or" or "and"; Requirements:

1. If the standard specification clause has subclauses, please determine the "or" or "and" relationship between the subclauses according to the content of the clause, and connect all the subclauses into a whole using "or" or "and"; 2. Please analyze the concluding sentence in the standard specification clause and place it at the end of the rewritten sentence in the form of "conclude: + concluding sentence"; 3. The punctuation marks in the clause that represent parallelism should also be analyzed to determine whether they serve as "or" or "and". Replace them with "or" or "and" if they meet the conditions. The parts separated by punctuation should be supplemented with the appropriate verbs to ensure grammatical correctness; The given sentence is "{input}", and only the modified sentence should be output in the output -------------”； When using the above prompt words to complete the clause, you only need to replace the {input} in the prompt words with the standard specification clause to be completed, and send the entire prompt to the natural language large model to get the completed standard specification clause.

5. The hybrid inference BIM design outcome auditing method of claim 3, wherein, The natural language large model prompt for decomposing the completed specification clause in S22 is as follows: "------------- Assume you are a standard specification clause semantic analysis expert; now define the AND relationship: all conditions must be met simultaneously; define the OR relationship: conditions are mutually exclusive, and meeting one is sufficient; analyze the semantics of the given standard specification clause, and split the standard specification clause into multiple independent clauses according to the OR relationship; Requirements:

1. Ensure that the constraints within each clause are connected by the OR relationship; 2. Ensure that each clause can independently and completely lead to the conclusion of the clause; The given clause is "{input}"; -------------”； When using the above prompt words to split the clause, replace {input} with the clause to be split to get the target clause.

6. The hybrid inference BIM design outcome auditing method of claim 3, wherein, The natural language large model prompt for converting clause subclauses into triples in S23 is as follows: "------------- Assume you are an entity relationship triple extraction model; first determine which entities (subject and object) and relationship types (relation) are in the text, then extract the subject and object from the sentence, and form a triple in the form (subject, relation, object); Define the AND relationship: all conditions must be met simultaneously; these conditions will be split into independent triples one by one, but they represent a whole fact when combined; when encountering the AND relationship: continue to split each condition to generate multiple triples, indicating that all conditions must be met simultaneously; Define the OR relationship: conditions are mutually exclusive, and meeting one is sufficient; These conditions will be split into different branches, each branch representing a possible case; encounter OR relationship: branch processing, each branch generates a set of triples, indicating different possibilities; The given sentence is "{input}", and only the final triple is output. -------------”； When using the above natural language large model prompt to convert standard specification clauses into triples, replace "{input}" with the clause to be converted, and obtain the final triples.

7. The hybrid inference BIM design outcome auditing method of claim 1, wherein, In S6, the rule reasoning audit is performed using the Jena reasoning engine. The GenericRuleReasoner class of the Apache Jena library is used to perform reasoning and audit on the BIM design results file in OWL format.

8. The hybrid inference BIM design outcome auditing method of claim 1, wherein, The natural language large model prompt for auditing the design results information and the original natural language form of the specification clauses in S6 is: "------------- You are a railway BIM audit expert, please audit the BIM design results information provided by me according to the standard specification clauses. If the design results information provided meets the requirements of the standard specification clauses, directly output "qualified", if not, point out the unqualified content and give the reason for unqualification; now, the standard specification is "{rule}", the BIM design results information provided is "{info}", what is your audit result? -------------”； When using the above prompt to audit the compliance of BIM design results, replace "{rule}" with the original natural language form of the specification clauses, and replace "{info}" with the design results information described in the Json text generated by S4, input the natural language large model, and obtain the reasoning result.

Images