CN117971904A - Building information model expert system and method based on large language model and attribute map - Google Patents
Building information model expert system and method based on large language model and attribute map Download PDFInfo
- Publication number
- CN117971904A CN117971904A CN202410362197.7A CN202410362197A CN117971904A CN 117971904 A CN117971904 A CN 117971904A CN 202410362197 A CN202410362197 A CN 202410362197A CN 117971904 A CN117971904 A CN 117971904A
- Authority
- CN
- China
- Prior art keywords
- node
- module
- information
- attribute
- knowledge
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000000605 extraction Methods 0.000 claims abstract description 68
- 230000006872 improvement Effects 0.000 claims abstract description 9
- 238000009960 carding Methods 0.000 claims abstract description 5
- 230000008520 organization Effects 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 13
- 238000010276 construction Methods 0.000 claims description 11
- 238000011161 development Methods 0.000 claims description 7
- 238000011156 evaluation Methods 0.000 claims description 6
- 239000000463 material Substances 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 4
- 230000036541 health Effects 0.000 claims description 4
- 238000012423 maintenance Methods 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 claims description 4
- 238000007726 management method Methods 0.000 description 39
- 238000010586 diagram Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 206010010219 Compulsions Diseases 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 239000011083 cement mortar Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000945 filler Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000004570 mortar (masonry) Substances 0.000 description 1
- 238000013439 planning Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a building information model expert system and a building information model expert method based on a large language model and an attribute map. It comprises the following steps: the expert knowledge information extraction and structuring module is used for carrying out structuring carding on unstructured or semi-structured expert knowledge in the building field through a large language model, and extracting attribute information of the expert knowledge as priori knowledge of reasoning and suggestion; the expert knowledge structured storage module is used for storing structured expert knowledge by using an attribute graph and ensuring the accuracy and the high efficiency of query; the building information extraction module is used for extracting building information from the building information model and taking the building information as the basis and the object of reasoning and suggestion; and the knowledge reasoning and suggestion module is used for inquiring and integrating the expert knowledge or the building information and outputting suggestions through the reasoning capability of the large language model. The invention can efficiently utilize expert knowledge in the whole life cycle of the building, and provide reasonable improvement suggestions according to the management current situation, thereby assisting in improving the management quality of the whole life cycle of the building.
Description
Technical Field
The invention relates to the field of intelligent construction, in particular to a building information model expert system and method based on a large language model and an attribute map.
Background
In the whole life cycle management process of a building based on a building information model, a great deal of expert knowledge such as building specifications, enterprise regulations, project data, personal experience and the like can be involved in each stage of design, planning, construction, supervision, transportation, dismantling and the like. These expert knowledge are the basis for guaranteeing the quality safety of the construction engineering. However, due to the extensive, fragmented and unstructured factors of the management and utilization modes, the expert knowledge is often difficult to be efficiently utilized and fully implemented, and an efficient query and management method is also lacking.
At present, knowledge engineering related technology based on artificial intelligence gradually shows potential in the professional field, and a specific form mainly comprises a knowledge question-answering system and a knowledge question-answering method. However, for the existing knowledge question-answering method, its practical performance is limited by the accuracy of extraction of the professional domain entity, the maximum length of the text vector, the limitation of RDF in describing complex behaviors, and the like. In addition, the traditional knowledge question-answering method cannot extract and utilize the building information contained in the building information model, so that the answer content of the knowledge question-answering method is separated from the reality of the project, and the suggestion cannot be specifically proposed.
Therefore, it is necessary to closely combine model information and expert knowledge for the characteristics of the building full life cycle management process, propose effective project improvement suggestions, and finally improve the building quality and the building full life cycle management level.
Disclosure of Invention
The invention aims to provide a building information model expert system and a building information model expert method based on a large language model and an attribute map, wherein expert knowledge is structured and arranged through the large language model, the attribute map and NoSQL are stored and managed, building information in the building information model is used as a basis and an object, the reasoning capacity of the large language model is stimulated, and finally effective project improvement suggestions are provided.
In order to achieve the above purpose, the invention adopts the following technical scheme:
The building information model expert system based on the large language model and the attribute map comprises an expert knowledge information extraction and structuring module, an expert knowledge structuring storage module, a building information extraction module and a knowledge reasoning and suggesting module; wherein,
The expert knowledge information extraction and structuring module is used for carrying out structuring carding on unstructured or semi-structured building field expert knowledge such as industry specifications, enterprise regulations and project documents and the like through a large language model, and comprises the organization structure of nodes, extraction of all levels of nodes and attributes thereof, unique identification and alignment of the nodes, relation connection among all levels of nodes and extraction of relation attributes, and the like, which are used as prior knowledge of follow-up reasoning and suggestion;
the expert knowledge structured storage module is used for storing structured expert knowledge by using an attribute graph and ensuring the accuracy and the high efficiency of query;
the building information extraction module is used for extracting building information of the current state in the whole life cycle management process of the building from the building information model, and taking the building information as a basis and an object of reasoning and suggestion;
And the knowledge reasoning and suggestion module is used for inquiring and integrating the expert knowledge and the building information, and reasoning and outputting effective suggestions through a large language model.
The expert knowledge information extraction and structuring module comprises a node and attribute extraction sub-module, a node unique identification and alignment sub-module, a relationship attribute extraction sub-module and a node connection sub-module; the node and attribute extraction submodule thereof are used for respectively extracting attributes of the input text according to the node organization structures of project management tasks, subtasks, control items, treatises and keywords by using the large language model which is subjected to context learning, and outputting the attributes in a JSON format; the node uniqueness identification and alignment submodule is used for combining the extracted project management task, subtask, control item and article node number to obtain a corresponding uniqueness identification as a basis for node alignment, and the keyword node uses attribute values of name attribute and type attribute as a basis for node alignment; the relation attribute extraction sub-module is used for extracting attributes of each node and relation by using a large language model subjected to context learning, including description, conditions, compulsory and the like, and outputting the attributes in a JSON format; the node connection sub-module is used for carrying out relation connection on each node according to an organization structure, wherein relation labels among project management tasks, sub-tasks, control items and treatises with hierarchical classification relations are upper and lower levels, if a reference relation exists among the project management tasks, sub-tasks, control items and treatises without the upper and lower levels, the relation labels are references, and the relation labels among the treatises and keyword nodes contained in the treatises are contained and contained.
The expert knowledge structured storage module stores five types of nodes of project management tasks, subtasks, control items, treatises and keywords and attribute values thereof and a mixed topological structure formed by the five types of relationships and attribute values thereof in a mode of attribute graphs, wherein the upper level, the lower level, the quotation, the inclusion and the inclusion are carried out on the five types of nodes and the upper level, the lower level, the quotation, the inclusion and the inclusion are carried out on the upper level, the lower level, the quotation and the lower level of the nodes.
The building information extraction module comprises a global state information extraction sub-module, a local state information extraction sub-module and a component state information extraction sub-module, and is used for calling a building information model document development interface and extracting building information influencing the global, local and detailed component states of the current engineering project in the whole life cycle management process of the building; the global state information extraction submodule is used for extracting global state information including but not limited to project stage, project date, meteorological condition, construction plan, operation and maintenance plan, energy use plan and the like; the local state information extraction submodule is used for extracting local state information including but not limited to regional functions, regional phases, regional plans, structural health monitoring and the like; the component state information extraction submodule is used for extracting component state information including, but not limited to, component type, component size, component material, component cost quality and the like.
The knowledge reasoning and suggesting module comprises a knowledge inquiring sub-module and a reasoning and outputting sub-module; the knowledge inquiry submodule is used for inquiring related information in the attribute graph according to input content, extracting information of a problem and related item global state information, local state information and component state information of the problem by a user by using a large language model which is subjected to contextual learning to obtain key information, inquiring the key information in the attribute graph by using NoSQL to obtain an initial node directly containing the key information, inquiring nodes referenced by the initial node according to a reference relation, inquiring the bar nodes connected with the initial node and the reference node according to the upper level, lower level and the contained relation, and outputting attribute values of all inquired bar node content attributes; the reasoning and outputting submodule is used for comparing and analyzing the current project state and expert knowledge by using the large language model according to the requirement of task prompt, taking the output content of the knowledge inquiry submodule as priori knowledge, taking the building information extracted by the building information extraction module as basis and object and taking the user requirement or suggestion as guide, and finally reasoning to obtain evaluation, risk or improvement suggestion.
The implementation method of the building information model expert system based on the large language model and the attribute map specifically comprises the following steps:
S1, carrying out structural carding on unstructured or semi-structured building field expert knowledge such as industry specifications, enterprise regulations and project documents through a large language model, and taking the unstructured or semi-structured building field expert knowledge as priori knowledge of follow-up reasoning and suggestion;
S1-1, according to the types and characteristics of expert knowledge to be extracted, respectively compiling corresponding prompt words and 2-5 high-quality few-sample learning examples for five types of nodes of project management tasks, subtasks, control items, treatises and keywords, transmitting the prompt words and the 2-5 high-quality few-sample learning examples to a large language model in a historical dialogue form, enabling a user dialogue to be the text to be extracted, enabling an assistant dialogue to be an extraction result conforming to an output format, extracting the name, number, type, content, citation and other required attributes of the five types of nodes, and outputting the attributes in a JSON format;
S1-2, combining the numbers of the extracted project management tasks, subtasks, control items and treaty nodes to obtain a corresponding unique identifier I n = In-1 + Nn serving as a basis for node alignment, wherein I is the unique identifier, I 0 = Null, N is a numbered attribute value of the node, N is a hierarchical classification relation level of the node, N = 1,2, 3 and 4 are sequentially taken from the total to the separated, and the keyword node uses attribute values of name attributes and type attributes of the keyword node as the basis for node alignment;
S1-3, namely compiling corresponding prompt words and 2-5 high-quality few-sample learning examples for upper-level, lower-level, quoted, contained and contained five types of relations, transmitting the prompt words and the 2-5 high-quality few-sample learning examples to a large language model in a historical dialogue form, enabling a user dialogue to be a text to be extracted, enabling an assistant dialogue to be an extraction result conforming to an output format, extracting required attributes such as description, conditions, compulsions and the like of the relations among all nodes, and outputting the extracted attributes in a JSON format;
s1-4, connecting all nodes according to an organization structure according to the obtained JSON object, wherein the relation labels among project management tasks, subtasks, control items and treatises with hierarchical classification relations are superior and inferior, if a reference relation exists among the project management tasks, subtasks, control items and treatises without superior and inferior relations, the relation labels are references, and the relation labels among the treatises and keyword nodes contained in the treatises are contained and included;
S2, storing a project management task, a subtask, a control item, a treaty, a keyword five-class node and attribute values thereof and a mixed topological structure formed by the upper level, the lower level, the reference, the inclusion and the inclusion of the five-class relationship and the attribute values thereof in a mode of an attribute map, and managing the project management task, the subtask, the control item, the treaty, the keyword five-class node and the attribute values thereof through NoSQL;
S3, calling a building information model document development interface, and extracting building information influencing the states of global, local and detailed components of the current engineering project in the whole life cycle management process of the building;
S3-1, extracting building information which is required by project phases, project dates, meteorological conditions, construction plans, operation and maintenance plans, energy use plans and the like and can influence the global state of the project;
s3-2, extracting building information which is required by regional functions, regional phases, regional plans, structural health monitoring and the like and can influence the local state of the project;
S3-3, extracting building information which is required by the types of the components, the sizes of the components, the materials of the components, the cost quality of the components and the like and can influence the states of the components of the project detail;
s4, inquiring and integrating expert knowledge and building information, and reasoning and outputting effective suggestions through a large language model;
S4-1, respectively compiling prompt words and 2-5 high-quality few-sample learning examples for project global state information, local state information and component state information, transmitting the prompt words and the 2-5 high-quality few-sample learning examples to a large language model in a historical dialogue form, enabling a user dialogue to be a text to be extracted, enabling an assistant dialogue to be an extraction result conforming to an output format, and then carrying out information extraction on a problem raised by a user and related project global state information, local state information and component state information thereof to obtain key information;
S4-2, inquiring the key information in the attribute graph by using NoSQL to obtain an initial node directly containing the key information, inquiring nodes referenced by the initial node according to the reference relation, and finally inquiring the treaty nodes connected with the initial node and the reference nodes thereof according to the superior, subordinate, contained and contained relation, and outputting attribute values of all inquired treaty node content attributes;
s4-3, compiling corresponding prompt words according to user requirements and preset task types, using a large language model to obtain evaluation, risk or improvement suggestions according to requirements of the prompt words, taking output content of the S4-2 as priori knowledge, taking building information extracted by the S3 as a basis and an object, taking the user requirements or suggestions as guidance, comparing and analyzing the current project state and expert knowledge, and finally reasoning.
Compared with the prior art, the invention has the following characteristics and beneficial effects:
(1) The invention improves the utilization level of expert knowledge such as building standards, enterprise regulations, project data and the like in the whole life cycle management process of the building based on the building information model by using the knowledge engineering and expert system method, and is beneficial to implementing related requirements in the whole process management of the building;
(2) The large language model and attribute map method and the mixed topological structure formed by the five-class nodes and the five-class relations thereof have strong readability and convenient inquiry, are not easy to be limited by the extraction capacity of model information and the length of text vectors, and are more suitable for engineering habits in organization and use;
(3) The invention extracts, analyzes and infers the building information such as global, local, detail and the like contained in the building information model in the whole life cycle management process, solves the problem that the traditional knowledge question-answering method can not acquire and utilize the actual building information, and can provide actual evaluation, risk or improvement suggestion for engineering for users.
Drawings
FIG. 1 is a diagram of a building information model expert system architecture based on a large language model and attribute map in accordance with the present invention;
FIG. 2 is a block diagram of a node and relationship attribute diagram used in the present embodiment;
FIG. 3 is a flow chart of a method of implementing the building information model expert system of the present invention based on a large language model and an attribute map.
Detailed Description
In order to make the objects, technical solutions and features of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and examples. Unless otherwise defined, all technical and scientific terms used hereinafter have the same meaning as commonly understood in the scientific and technical arts to which this invention belongs. The specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the invention.
FIG. 1 illustrates an example of a building information model expert system based on a large language model and an attribute map, comprising an expert knowledge information extraction and structuring module, an expert knowledge structured storage module, a building information extraction module and a knowledge reasoning and suggesting module;
The expert knowledge information extraction and structuring module is used for carrying out structuring carding on unstructured or semi-structured building field expert knowledge such as industry specifications, enterprise regulations and project documents and the like through a large language model, and comprises the organization structure of nodes, extraction of all levels of nodes and attributes thereof, unique identification and alignment of the nodes, relation connection among all levels of nodes and extraction of relation attributes, and the like, which are used as prior knowledge of follow-up reasoning and suggestion;
the expert knowledge structured storage module is used for storing structured expert knowledge by using an attribute graph and ensuring the accuracy and the high efficiency of query;
the building information extraction module is used for extracting building information of the current state in the whole life cycle management process of the building from the building information model, and taking the building information as a basis and an object of reasoning and suggestion;
And the knowledge reasoning and suggestion module is used for inquiring and integrating the expert knowledge and the building information, and reasoning and outputting effective suggestions through a large language model.
The expert knowledge information extraction and structuring module comprises a node and attribute extraction sub-module, a node unique identification and alignment sub-module, a relationship attribute extraction sub-module and a node connection sub-module; the node and attribute extraction submodule thereof are used for respectively extracting attributes of the input text according to the node organization structures of project management tasks, subtasks, control items, treatises and keywords by using the large language model which is subjected to context learning, and outputting the attributes in a JSON format; the node uniqueness identification and alignment submodule is used for combining the extracted project management task, subtask, control item and article node number to obtain a corresponding uniqueness identification as a basis for node alignment, and the keyword node uses attribute values of name attribute and type attribute as a basis for node alignment; the relation attribute extraction sub-module is used for extracting attributes of each node and relation by using a large language model subjected to context learning, including description, conditions, compulsory and the like, and outputting the attributes in a JSON format; the node connection sub-module is used for carrying out relation connection on each node according to an organization structure, wherein relation labels among project management tasks, sub-tasks, control items and treatises with hierarchical classification relations are upper and lower levels, if a reference relation exists among the project management tasks, sub-tasks, control items and treatises without the upper and lower levels, the relation labels are references, and the relation labels among the treatises and keyword nodes contained in the treatises are contained and contained. The present embodiment uses, but is not limited to, a lightweight large language model ChatGLM-6B, and the structure of the property graph and the level of nodes, relationships, and their property settings are shown in fig. 2.
The expert knowledge structured storage module stores five types of nodes and attribute values of the five types of nodes and superior, subordinate, quoted, contained and contained relationships among the nodes of project management tasks, subtasks, control items, treatises and keywords and a mixed topology structure formed by the five types of relationships and the attribute values thereof in a mode of an attribute map, and manages the mixed topology structure through NoSQL. This embodiment employs, but is not limited to Neo4j 5.18.0 and Cypher Query Language as a storage and management tool for attribute maps.
The building information extraction module comprises a global state information extraction sub-module, a local state information extraction sub-module and a component state information extraction sub-module, and is used for calling a building information model document development interface and extracting building information influencing the global, local and detailed component states of the current engineering project in the whole life cycle management process of the building. The global state information extraction submodule is used for extracting global state information including but not limited to project stage, project date, meteorological condition, construction plan, operation and maintenance plan, energy use plan and the like; the local state information extraction submodule is used for extracting local state information including but not limited to regional functions, regional phases, regional plans, structural health monitoring and the like; the component state information extraction submodule is used for extracting component state information including, but not limited to, component type, component size, component material, component cost quality and the like. The present embodiment uses, but is not limited to, revit2024 as a building information model creation and management platform, and RevitPythonShell as a model document invocation and development tool.
The knowledge reasoning and suggesting module comprises a knowledge inquiring sub-module and a reasoning and outputting sub-module; the knowledge inquiry submodule is used for inquiring related information in the attribute graph according to input content, extracting information of a problem and related item global state information, local state information and component state information of the problem by a user by using a large language model which is subjected to contextual learning to obtain key information, inquiring the key information in the attribute graph by using NoSQL to obtain an initial node directly containing the key information, inquiring nodes referenced by the initial node according to a reference relation, inquiring the bar nodes connected with the initial node and the reference node according to the upper level, lower level and the contained relation, and outputting attribute values of all inquired bar node content attributes; the reasoning and outputting submodule is used for comparing and analyzing the current project state and expert knowledge by using the large language model according to the requirement of task prompt, taking the output content of the knowledge inquiry submodule as priori knowledge, taking the building information extracted by the building information extraction module as basis and object and taking the user requirement or suggestion as guide, and finally reasoning to obtain evaluation, risk or improvement suggestion.
Fig. 3 shows an example of a method for implementing the building information model expert system based on a large language model and an attribute map, which specifically includes the following steps:
Step one: unstructured or semi-structured building domain expert knowledge such as industry specifications, enterprise regulations and project documents is structurally carded through a large language model to serve as priori knowledge of follow-up reasoning and suggestion, and the embodiment adopts, but is not limited to, the following table 1 data as original expert knowledge;
TABLE 1
，
1. Respectively compiling corresponding prompt words and 2-5 high-quality few-sample learning examples for five types of nodes of project management tasks, subtasks, control items, treatises and keywords, extracting information of expert knowledge by using a large language model, and outputting the information in a JSON format; in order to improve the stability of output, one of parameters such as temperature, top _k or top_p can be selected to be reduced as much as possible;
2. The numbers of the five types of nodes are combined to obtain unique identifiers thereof; the key word nodes take attribute values of name attributes and type attributes thereof as the basis of node alignment, and other four types of nodes take unique identifiers as the basis of alignment; for example, five types of nodes centered on the first piece of content can be output as:
{ tag: project management tasks, attributes: { name: masonry structure management scheme, ID: QT-001},
{ Tag: subtasks, attributes: { name: construction requirements, ID: QT-001-1},
{ Tag: control item, attribute: { name: frame infill wall, ID: QT-001-1-1.1},
{ Tag: treaty, attribute: { ID: QT-001-1-1.1-1.1.1, content: the filler wall length exceeds …,
{ Tag: keywords, attributes: { name: filling wall, constructional column ], type: [ wall, column ] };
3. The method comprises the steps of respectively compiling corresponding prompt words and 2-5 high-quality few-sample learning examples for upper-level, lower-level, quoted and contained five types of relations, extracting required attributes such as description, conditions, compulsory and the like of the relations among all nodes, and outputting the extracted attributes in a JSON format;
4. and connecting all nodes according to the organization structure according to the obtained JSON object, wherein the relation labels among the project management tasks, the subtasks, the control items and the treatises with hierarchical classification relations are upper and lower levels, and if a reference relation exists among the project management tasks, the subtasks, the control items and the treatises without upper and lower levels, the relation labels are references, and the relation labels among the treatises and the keyword nodes contained in the treatises are contained and contained.
Step two: and inputting the project management task, subtasks, control items, treatises, keyword five-class nodes and attribute values thereof and all information among the nodes, including upper level, lower level, reference, included and included five-class relations and attribute values thereof into neo4j, creating nodes and relations to generate an attribute map, and carrying out management and subsequent query operation through Cypher Query Language.
Step three: calling a building information model document development interface, and extracting building information influencing the states of global, local and detailed components of a current engineering project in the whole life cycle management process of the building;
1. The building information model document is obtained, global parameters of all projects are obtained by using GetAllGlobalParameters and other methods, and the global parameters extracted in the embodiment are as follows: date: 12 month xx day, weather: -6-5 ℃, sunny stage: masonry engineering construction ";
2. the building information model document is obtained, the required local parameters are obtained by using an element. Parameters method, etc., and the parameters extracted to a certain room in the embodiment are as follows: category: room, name: bathroom, elevation: 6.000, stage: masonry construction preparation ";
3. The building information model document is obtained, the required component parameters are obtained by using an element. Parameters method, etc., and the parameters of a certain masonry infill wall in the room are extracted in the embodiment, for example: category: wall, type: infill wall, length: 6m, height: 3m, elevation: 6.000, material: autoclaved aerated concrete blocks, M10 cement mortar.
Step four: inquiring and integrating expert knowledge and building information, and reasoning and outputting effective suggestions through a large language model;
1. Compiling prompt words and 2-5 high-quality few-sample learning examples for the project global state information, the local state information and the component state information respectively, and then extracting information of the project global state information, the local state information and the component state information related to the problems proposed by the user to obtain key information; the key information extracted for query in this embodiment is: "[ toilet, infilled wall, autoclaved aerated concrete block, mortar ]";
2. By using Cypher Query Language to query the extracted key information, the key word node with the corresponding name can be queried, the corresponding treaty node is queried through the containing or contained relation, the attribute value of the content attribute of the treaty node is queried, and 5 pieces of expert knowledge shown in table 1 are obtained;
3. According to the user requirement and the preset task type, a corresponding prompt word is compiled, a large language model is used for generating a recommendation which is finally inferred and output by the embodiment according to the requirement of the prompt word, the obtained 5 pieces of expert knowledge are taken as priori knowledge, and three building information including the global parameter, the certain room parameter and the certain masonry infill wall parameter extracted in the step three are taken as the basis and the object of concrete recommendation, wherein the recommendation is as follows: the middle part of the wall body is provided with a constructional column, the bottom of the wall body is provided with a high concrete sill with 150mm cast-in-place, the autoclaved aerated concrete block is not wetted by water, and the gap between the filling wall and the bearing main body structure is constructed after the filling wall is built for 14 d;
4. It should be noted that, the inference effect and the specific output result will be different due to different model performance, parameter setting, prompt words, few sample learning examples, etc., but the expected requirement can be satisfied after a small amount of adaptive adjustment.
The foregoing description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. It should be noted that it will be apparent to those skilled in the art that modifications and substitutions can be made to some of the features of the present invention without departing from the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.
Claims (8)
1. The building information model expert system based on the large language model and the attribute map is characterized by comprising an expert knowledge information extraction and structuring module, an expert knowledge structuring storage module, a building information extraction module and a knowledge reasoning and suggesting module; wherein,
The expert knowledge information extraction and structuring module is used for carrying out structuring carding on unstructured or semi-structured expert knowledge in the building field through a large language model, and comprises the organization structure of nodes, extraction of all levels of nodes and attributes thereof, unique identification and alignment of the nodes, relation connection between all levels of nodes and extraction of relation attributes, wherein the relation connection and the extraction of the relation attributes are used as priori knowledge of follow-up reasoning and suggestion;
the expert knowledge structured storage module is used for storing structured expert knowledge by using an attribute graph and ensuring the accuracy and the high efficiency of query;
the building information extraction module is used for extracting building information of the current state in the whole life cycle management process of the building from the building information model, and taking the building information as a basis and an object of reasoning and suggestion;
And the knowledge reasoning and suggestion module is used for inquiring and integrating the expert knowledge and the building information, and reasoning and outputting effective suggestions through a large language model.
2. The building information model expert system based on a large language model and an attribute map according to claim 1, wherein the expert knowledge information extraction and structuring module comprises a node and its attribute extraction sub-module, a node unique identification and alignment sub-module, a relationship attribute extraction sub-module and a node connection sub-module;
The node and attribute extraction submodule thereof are used for respectively extracting attributes of the input text according to the node organization structures of project management tasks, subtasks, control items, treatises and keywords by using the large language model subjected to the context learning, and outputting the input text in a JSON format;
the node uniqueness identification and alignment submodule is used for combining the extracted project management task, subtask, control item and article node number to obtain a corresponding uniqueness identification as a basis for node alignment, and the keyword nodes use attribute values of name attributes and type attributes as the basis for node alignment;
the relation attribute extraction sub-module is used for extracting attributes of each node and each relation by using a large language model subjected to context learning, and outputting the attributes in a JSON format;
The node connection sub-module is used for carrying out relation connection on each node according to an organization structure, wherein relation labels among project management tasks, sub-tasks, control items and treatises with hierarchical classification relations are upper and lower levels, if a reference relation exists among the project management tasks, sub-tasks, control items and treatises without upper and lower levels, the relation labels are references, and the relation labels among the treatises and keyword nodes contained in the treatises are contained and contained.
3. The large language model and attribute map based building information model expert system of claim 2 wherein the extracting attributes of nodes and relationships using the context learned large language model and outputting in JSON format comprises:
The prompting word for the context learning comprises three parts of contents including task description, task requirement and format requirement;
2-5 high-quality few-sample learning examples are provided based on a historical dialogue mode, a user dialogue is text to be extracted, and an assistant dialogue is an extraction result conforming to an output format.
4. The large language model and attribute map based building information model expert system of claim 2 wherein the combining of extracted project management tasks, subtasks, control items and treaty node numbers comprises:
The node unique identifier I n = In-1 + Nn, wherein I is a unique identifier, I 0 =null, N is a node number attribute value, N is a node hierarchical classification relation level, and n=1, 2,3 and 4 are sequentially taken from the total score.
5. The large language model and attribute map based building information model expert system of claim 2 wherein the expert knowledge structured storage module stores project management tasks, subtasks, control items, treatises, keyword five classes of nodes and their attribute values and upper, lower, reference, include and are comprised of a hybrid topology of five classes of relationships and their attribute values in an attribute map manner and is managed by NoSQL.
6. The large language model and attribute map based building information model expert system of claim 5 wherein the building information extraction module includes a global status information extraction sub-module, a local status information extraction sub-module, and a component status information extraction sub-module for invoking a building information model document development interface to extract building information affecting global, local, and detailed component status of a current project during a full life cycle management of a building.
7. The large language model and attribute map based building information model expert system of claim 6 wherein the knowledge reasoning and suggestion module includes a knowledge query sub-module and a reasoning and output sub-module;
The knowledge query submodule is used for querying relevant information in the attribute graph according to input content, and comprises the following steps:
Extracting information of a problem and related project global state information, local state information and component state information of the problem by a user by using a large language model through contextual learning to obtain key information, inquiring the key information by using NoSQL in an attribute graph to obtain an initial node directly containing the key information, inquiring a node quoted by the initial node according to a quoted relation, and finally inquiring the initial node and the treaty node connected with the quoted node according to a superior, subordinate and contained relation, and outputting attribute values of content attributes of all inquired treaty nodes;
The reasoning and outputting submodule is used for comparing and analyzing the current project state and expert knowledge by using the large language model according to the requirement of task prompt, taking the output content of the knowledge inquiry submodule as priori knowledge, taking the building information extracted by the building information extraction module as basis and object and taking the user requirement or suggestion as guide, and finally reasoning to obtain evaluation, risk or improvement suggestion.
8. The implementation method of the building information model expert system based on the large language model and the attribute map is characterized by comprising the following steps:
S1, carrying out structural combing on unstructured or semi-structured expert knowledge in the construction field through a large language model, and taking the structured combing as prior knowledge of follow-up reasoning and suggestion;
S1-1, according to the types and characteristics of expert knowledge to be extracted, respectively compiling corresponding prompt words and 2-5 high-quality few-sample learning examples for five types of nodes of project management tasks, subtasks, control items, treatises and keywords, transmitting the prompt words and the 2-5 high-quality few-sample learning examples to a large language model in a historical dialogue form, enabling a user dialogue to be the text to be extracted, enabling an assistant dialogue to be an extraction result conforming to an output format, extracting names, numbers, types, contents and reference attributes of the five types of nodes, and outputting the names, the numbers, the types, the contents and the reference attributes in a JSON format;
S1-2, combining the numbers of the extracted project management tasks, subtasks, control items and treaty nodes to obtain a corresponding unique identifier I n = In-1 + Nn serving as a basis for node alignment, wherein I is the unique identifier, I 0 = Null, N is a numbered attribute value of the node, N is a hierarchical classification relation level of the node, N = 1,2, 3 and 4 are sequentially taken from the total to the separated, and the keyword node uses attribute values of name attributes and type attributes of the keyword node as the basis for node alignment;
S1-3, namely compiling corresponding prompt words and 2-5 high-quality few-sample learning examples for upper-level, lower-level, quoted, contained and contained five types of relations, transmitting the prompt words and the 2-5 high-quality few-sample learning examples to a large language model in a historical dialogue form, enabling a user dialogue to be a text to be extracted, enabling an assistant dialogue to be an extraction result conforming to an output format, extracting description, conditions and mandatory properties of relations among nodes, and outputting the extracted results in a JSON format;
s1-4, connecting all nodes according to an organization structure according to the obtained JSON object, wherein the relation labels among project management tasks, subtasks, control items and treatises with hierarchical classification relations are superior and inferior, if a reference relation exists among the project management tasks, subtasks, control items and treatises without superior and inferior relations, the relation labels are references, and the relation labels among the treatises and keyword nodes contained in the treatises are contained and included;
S2, storing a project management task, a subtask, a control item, a treaty, a keyword five-class node and attribute values thereof and a mixed topological structure formed by the upper level, the lower level, the reference, the inclusion and the inclusion of the five-class relationship and the attribute values thereof in a mode of an attribute map, and managing the project management task, the subtask, the control item, the treaty, the keyword five-class node and the attribute values thereof through NoSQL;
S3, calling a building information model document development interface, and extracting building information influencing the states of global, local and detailed components of the current engineering project in the whole life cycle management process of the building;
S3-1, extracting global state information comprising project stage, project date, meteorological condition, construction plan, operation and maintenance plan and energy use plan;
s3-2, extracting local state information comprising regional functions, regional phases, regional plans and structural health monitoring;
s3-3, extracting component state information comprising component type, component size, component material and component cost quality;
s4, inquiring and integrating expert knowledge and building information, and reasoning and outputting effective suggestions through a large language model;
S4-1, respectively compiling prompt words and 2-5 high-quality few-sample learning examples for project global state information, local state information and component state information, transmitting the prompt words and the 2-5 high-quality few-sample learning examples to a large language model in a historical dialogue form, enabling a user dialogue to be a text to be extracted, enabling an assistant dialogue to be an extraction result conforming to an output format, and then carrying out information extraction on a problem raised by a user and related project global state information, local state information and component state information thereof to obtain key information;
S4-2, inquiring the key information in the attribute graph by using NoSQL to obtain an initial node directly containing the key information, inquiring nodes referenced by the initial node according to the reference relation, and finally inquiring the treaty nodes connected with the initial node and the reference nodes thereof according to the superior, subordinate, contained and contained relation, and outputting attribute values of all inquired treaty node content attributes;
s4-3, compiling corresponding prompt words according to user requirements and preset task types, using a large language model to obtain evaluation, risk or improvement suggestions according to requirements of the prompt words, taking output content of the S4-2 as priori knowledge, taking building information extracted by the S3 as a basis and an object, taking the user requirements or suggestions as guidance, comparing and analyzing the current project state and expert knowledge, and finally reasoning.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410362197.7A CN117971904B (en) | 2024-03-28 | 2024-03-28 | Building information model expert system and method based on large language model and attribute map |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410362197.7A CN117971904B (en) | 2024-03-28 | 2024-03-28 | Building information model expert system and method based on large language model and attribute map |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117971904A true CN117971904A (en) | 2024-05-03 |
| CN117971904B CN117971904B (en) | 2024-07-05 |
Family
ID=90859745
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410362197.7A Active CN117971904B (en) | 2024-03-28 | 2024-03-28 | Building information model expert system and method based on large language model and attribute map |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117971904B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5019961A (en) * | 1989-04-05 | 1991-05-28 | Cadware, Inc. | Computer apparatus and method for logical modelling |
| WO1996022575A1 (en) * | 1995-01-17 | 1996-07-25 | Intertech Ventures, Ltd. | Control systems based on simulated virtual models |
| CN111666622A (en) * | 2020-06-11 | 2020-09-15 | 厦门海迈科技股份有限公司 | Method and device for constructing BIM full-period data model |
| CN112116713A (en) * | 2020-09-19 | 2020-12-22 | 南昌大学 | High-precision steel bar arrangement method for linear bearing platform type components |
| CN112784346A (en) * | 2021-02-07 | 2021-05-11 | 殿汇空间(上海)信息科技有限公司 | Building structure autonomous design method, system, terminal and storage medium |
| CN113434928A (en) * | 2021-05-26 | 2021-09-24 | 南昌大学 | Parametric construction method of complex three-dimensional linear structure |
| CN114444180A (en) * | 2022-01-18 | 2022-05-06 | 河北工业大学 | Full life cycle parameter prediction and monitoring method and system for assembly type building structure |
-
2024
- 2024-03-28 CN CN202410362197.7A patent/CN117971904B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5019961A (en) * | 1989-04-05 | 1991-05-28 | Cadware, Inc. | Computer apparatus and method for logical modelling |
| WO1996022575A1 (en) * | 1995-01-17 | 1996-07-25 | Intertech Ventures, Ltd. | Control systems based on simulated virtual models |
| CN111666622A (en) * | 2020-06-11 | 2020-09-15 | 厦门海迈科技股份有限公司 | Method and device for constructing BIM full-period data model |
| CN112116713A (en) * | 2020-09-19 | 2020-12-22 | 南昌大学 | High-precision steel bar arrangement method for linear bearing platform type components |
| CN112784346A (en) * | 2021-02-07 | 2021-05-11 | 殿汇空间(上海)信息科技有限公司 | Building structure autonomous design method, system, terminal and storage medium |
| CN113434928A (en) * | 2021-05-26 | 2021-09-24 | 南昌大学 | Parametric construction method of complex three-dimensional linear structure |
| CN114444180A (en) * | 2022-01-18 | 2022-05-06 | 河北工业大学 | Full life cycle parameter prediction and monitoring method and system for assembly type building structure |
Non-Patent Citations (2)
| Title |
|---|
| L. C. M. TANG 等: "Intelligent BVAC information capturing system for smart building information modelling", 2013 5TH INTERNATIONAL CONFERENCE ON POWER ELECTRONICS SYSTEMS AND APPLICATIONS(PESA), 12 June 2014 (2014-06-12) * |
| 李丹宇 等: "智能建筑设计的发展研究与探索", 中国建筑装饰装修, 5 December 2023 (2023-12-05) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117971904B (en) | 2024-07-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Visser et al. | Ontologies for geographic information processing | |
| Li et al. | Learning ontology from relational database | |
| Krishnan | PDM: A knowledge-based tool for model construction | |
| Smirnov et al. | Domain ontologies integration for virtual modelling and simulation environments | |
| CN115757810A (en) | Method for constructing standard ontology of knowledge graph | |
| Bundy | Intelligent front ends | |
| CN114117065A (en) | Knowledge graph construction method and system based on power production statistical service | |
| CN112307772B (en) | Construction method of broad-color porcelain knowledge base based on semantic ontology | |
| Xue et al. | Frame-based ontological view for semantic integration | |
| Avdeenko et al. | The ontology-based approach to support the requirements engineering process | |
| CN117971904B (en) | Building information model expert system and method based on large language model and attribute map | |
| Jiang et al. | Construction of substation engineering design knowledge graph based on “ontology seven-step method” | |
| Aksamija et al. | Architectural ontology: development of machine-readable representations for building design drivers | |
| CN102270190A (en) | Computer modeling and solving processing method of complex decision-making problem | |
| Van Ban et al. | Querying Data with Fuzzy Information in Object-Oriented Databases Based on Hedge Algebraic Semantic | |
| Luo et al. | Ontology modeling method applied in simulation modeling of distribution network time series operation | |
| Zhang et al. | Transformation of transportation data models from unified modeling language to web ontology language | |
| Park et al. | RASES: A database supported framework for structured model base management | |
| CN114090790B (en) | Man-machine friendly data logic fusion power knowledge graph and construction method thereof | |
| Smirnov et al. | Operational decision support: Context-based approach and technological framework | |
| Srinivasan et al. | An object relational approach for the design of decision support systems | |
| Heavey et al. | A model management systems approach to manufacturing systems design | |
| Mekhilef et al. | AN UML MODELING OF AN ARCHITECTURE FOR KNOWLEDGE DOCUMENTATION | |
| Grossmann et al. | An Object-Oriented Simulation Environment Supporting Self-Adapting Model Structure and Qualitative Models | |
| Chevalier et al. | Wikibridge: A semantic wiki for archaeological applications |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |