Large and small model collaborative building completion digital delivery method

Abstract

According to the building completion digital delivery method based on cooperation of the large model and the small model, a large-scale language model and a series of small-scale professional models are trained, delivery information is provided for a user through cooperative work, and useful information contained in delivery data is truly applied. Through cooperative work of the large model and a plurality of professional small models, the building completion delivery information is intelligently processed, and the information processing efficiency, the retrieval precision and the personalized service capability are improved. The large model is responsible for professional knowledge understanding and complex information processing of the building industry, and the small model focuses on efficient processing of specific tasks. According to the method, various complex data formats in the building completion delivery process can be effectively processed through cooperative work of the large model and the small model, and the comprehensiveness and accuracy of information processing are improved. According to the method, required information can be quickly extracted from mass data through division and cooperation of the large and small models, and the method can be flexibly applied in different scenes, so that the information retrieval efficiency is greatly improved.

Description

Technical Field

[0001] The invention relates to a digital delivery method for building completion in collaboration with large and small models. Background Art

[0002] The number and scale of public buildings continue to expand, and the amount of information on completion and delivery is also increasing accordingly. The information contained in the delivery data includes detailed records of multiple links such as building design, construction, equipment installation and commissioning, which is crucial for subsequent operation and maintenance management. However, the current amount of building completion and delivery information is large, the format is not uniform, and it is difficult to retrieve, which makes it difficult to apply the information, seriously affecting the effective use of delivery information by the operation and maintenance team, and thus hindering the efficient operation and maintenance of building facilities. Summary of the invention

[0003] The object of the present invention is to provide a digital delivery method for building completion in which large and small models are coordinated.

[0004] To solve the above problems, the present invention provides a digital delivery method for building completion with large and small models in collaboration, comprising:

[0005] Collect and organize a large amount of documents and data in the construction industry to form the basic corpus T1;

[0006] Based on the special information of the completion and delivery stage of the current project, a professional corpus T2 for deep learning training is formed;

[0007] Based on the constructed basic corpus T1, train the large model M1 in the vertical field;

[0008] Train multiple small-scale neural network models M2 based on professional corpus T2;

[0009] A large model M1 and multiple small-scale neural network models M2 are used to collaboratively collect and analyze basic data;

[0010] The large model M1 and multiple small-scale neural network models M2 are used to collaborate and process the core tasks based on the delivered information;

[0011] A large model M1 and multiple small-scale neural network models M2 are used to collaborate to aggregate information and generate application responses.

[0012] Furthermore, in the above method, a large amount of documents and data in the construction industry are collected and sorted to form a basic corpus T1, including:

[0013] Collect and organize a large amount of documents and data in the construction industry, perform text recognition, document tagging, document content association and analysis on architectural design drawings, construction records, equipment installation documents, and technical specifications, and form a basic corpus T1 through preprocessing.

[0014] Furthermore, in the above method, a professional corpus T2 for deep learning training is formed for the special information of the completion and delivery stage that is unique to the current project, including:

[0015] Automated annotation and keyword extraction are performed on the special information of the completion and delivery stage that is unique to the current project to form a professional corpus T2 for deep learning training; the professional corpus T2 is used to store key information of a single project in the process of building completion and delivery, providing a data basis for the training of the small-scale neural network model M2.

[0016] Furthermore, in the above method, based on the constructed basic corpus T1, a large model M1 in a vertical field is trained, including:

[0017] The task of the big model M1 is to understand the professional knowledge of the construction industry, process documents in various formats, perform information extraction, document content analysis, knowledge question and answer, and code generation. The big model M1 is responsible for processing general tasks through a deep understanding of the semantics and context of the construction industry.

[0018] Furthermore, in the above method, multiple small-scale neural network models M2 are trained based on the professional corpus T2, including:

[0019] Based on the professional corpus T2, multiple small-scale neural network models M2 are trained. Each small-scale neural network model M2 is used to perform a specific small task. Through the execution of each small task, specific problems in the completion and delivery of the building can be quickly processed, and the processed data can be converted into structured information. The structured information is written into the knowledge base for subsequent query and retrieval by the large model M1.

[0020] Furthermore, in the above method, the small tasks include: image recognition and restoration, CAD file content extraction, point cloud data processing, image detection and problem identification, and path planning and task allocation.

[0021] Furthermore, in the above method, a large model M1 and multiple small-scale neural network models M2 are used to collaboratively collect and analyze basic data, including:

[0022] Receive the user's natural language query through the large model M1 and understand the query intent;

[0023] The large model M1 decomposes the query intent into specific query tasks and assigns them to the corresponding small-scale neural network model M2 according to the content of the query tasks;

[0024] Based on the query task, the small-scale neural network model M2 further analyzes the query intent and generates several specialized query subtasks in combination with the specific content of this engineering project.

[0025] Furthermore, in the above method, the query subtask includes: basement engineering image recognition and third floor ceiling quality problem detection.

[0026] Furthermore, in the above method, a large model M1 and multiple small-scale neural network models M2 are used in collaboration to deliver information to process core tasks, including:

[0027] The large model M1 searches for industry-level common texts or conducts large-scale searches and retrievals in massive amounts of delivery information documents based on query requirements;

[0028] The small-scale neural network model M2 processes the core tasks of a specific type of data related to this project, focusing on the delivery information of this project, locating the search objects for specific problems, and completing query subtasks.

[0029] Furthermore, in the above method, a large model M1 and multiple small-scale neural network models M2 are used to collaborate, aggregate information and generate application responses, including:

[0030] After each small-scale neural network model M2 completes the query subtask, it converts the result into the conclusion in the structured format required by the large model M1 and sends it back to the large model M1;

[0031] The large model M1 receives the results processed by the small model M2, aggregates the conclusions of all small models, organizes them into friendly natural language, and provides query results for users.

[0032] Compared with the prior art, the present invention proposes a method of using large language model technology to train a larger-scale language model and a series of smaller-scale professional models, which work together to provide users with delivery information and truly apply the useful information contained in the delivery materials. The present invention intelligently processes building completion delivery information through the collaborative work of a large model and multiple small professional models, thereby improving information processing efficiency, retrieval accuracy, and personalized service capabilities. The large model is responsible for understanding professional knowledge and complex information processing in the construction industry, while the small model focuses on efficient processing of specific tasks. The collaborative work of large and small models of the present invention can effectively handle a variety of complex data formats in the process of building completion delivery, and improve the comprehensiveness and accuracy of information processing. Through the division of labor and cooperation between large and small models, the present invention can quickly extract the required information from massive data and flexibly apply it in different scenarios, greatly improving the efficiency of information retrieval. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flowchart of a method for digital delivery of building completion in collaboration of large and small models according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] like Figure 1 As shown, the present invention provides a digital delivery method for building completion with coordinated large and small models, comprising:

[0036] Step S1, construction of delivery information corpus;

[0037] Step S1.1, construction of large model basic corpus:

[0038] Collect and organize a large amount of documents and data in the construction industry, perform text recognition, document tagging, document content association and analysis on data such as architectural design drawings, construction records, equipment installation documents, technical specifications, etc., and form a basic corpus T1 through preprocessing. For example, text extraction or recognition is performed on architectural design drawings, construction records, equipment installation documents, technical specifications, etc., the content of these documents is extracted, and the relevance is recorded.

[0039] For example, a large public library delivered a large amount of architectural drawings, equipment installation documents, maintenance manuals, etc. upon completion. Managers need to efficiently query equipment repair and maintenance information from the massive amount of delivery information.

[0040] The library collected architectural drawings, equipment installation documents, maintenance manuals and other materials, and digitized these documents through text recognition, document tagging and content association analysis to build a basic corpus T1. The basic corpus T1 includes information such as the model, installation location, and maintenance cycle of various types of equipment.

[0041] Step S1.2, small model basic corpus construction:

[0042] For the special information of the completion and delivery stage unique to the current project, automatic annotation, keyword extraction and other processing are performed to form a professional corpus T2 for deep learning training; the professional corpus T2 is used to store the key information of a single project in the process of building completion and delivery, and provide a data basis for the training of the small-scale neural network model M2. Preferably, the special information of the completion and delivery stage unique to the current project includes: the project's completion report, CAD files, PDF drawing scans, etc.

[0043] Specifically, for example, automatic annotation and keyword extraction are performed on library-specific equipment, such as automatic book lending and returning machines, air-conditioning systems, security monitoring systems, etc., to form a professional corpus T2, which provides a data basis for the training of the small-scale neural network model M2.

[0044] Step S2, training of large and small models.

[0045] Step S2.1, train a larger model M1:

[0046] Based on the constructed basic corpus T1, the large model M1 in the vertical field is trained; the task of the large model M1 is to understand the professional knowledge of the construction industry, and be able to process documents in various formats to perform tasks such as information extraction, document content analysis, knowledge question and answer, and code generation; the large model M1 is responsible for the processing of general tasks through a deep understanding of the semantics and context of the construction industry.

[0047] Specifically, for example, based on the basic corpus T1, a large model M1 in the vertical field is trained to enable it to understand the general knowledge of library buildings and equipment and perform tasks such as information extraction and document content analysis.

[0048] Step S2.2, training multiple smaller-scale professional large models as the small-scale neural network model M2:

[0049] Based on the professional corpus T2, multiple small-scale neural network models M2 are trained. Each small-scale neural network model M2 is used to perform specific small tasks. Through the execution of each small task, specific problems in the completion and delivery of the building are quickly processed, and the processed data is converted into structured information, and the structured information is written into the knowledge base for subsequent query and retrieval by the large model M1. Preferably, the small-scale neural network model M2 can perform specific small tasks such as image recognition and repair, CAD file content extraction, point cloud data processing, image detection and problem identification, path planning and task allocation.

[0050] Specifically, based on the professional corpus T2, multiple small models M2 are trained, such as the automatic book lending and returning machine maintenance model, the air conditioning system maintenance model, the security monitoring system maintenance model, etc., focusing on the maintenance and maintenance information processing of specific equipment.

[0051] Step S3, application of delivery information for collaborative large and small models.

[0052] Step S3.1, using the large model M1 and multiple small-scale neural network models M2 to collaboratively collect and analyze basic data:

[0053] First, the user’s natural language query is received through the large model M1 to understand the query intent;

[0054] Then, the large model M1 decomposes the query intent into specific query tasks and assigns them to the corresponding small-scale neural network model M2 according to the content of the query tasks;

[0055] Finally, the small-scale neural network model M2 further analyzes the query intent based on the query task, and generates several specialized query subtasks in combination with the specific content of the project. Preferably, the query subtasks include: basement engineering image recognition, third floor ceiling quality problem detection, etc.

[0056] Specifically, for example, when the library manager raises the query requirement "maintenance information of the central air-conditioning equipment on the third floor of the library", the large model M1 first receives and understands the query intent, and then decomposes the intent into specific tasks and assigns them to the corresponding small model M2.

[0057] Step S3.2, using the large model M1 and multiple small-scale neural network models M2 to collaborate and process the core tasks based on the delivered information:

[0058] The large model M1 searches for industry-level common texts or conducts large-scale searches and retrievals in massive amounts of delivery information documents based on query requirements;

[0059] The small-scale neural network model M2 processes the core task of a specific type of data related to this project, focusing on the delivery information of this project, locating the search object of the specific problem, and completing the query subtask;

[0060] Specifically, for example, after receiving a query request, the large model M1 is responsible for conducting a large-scale search and retrieval in a massive amount of delivery information documents. The large model M1 will first find relevant general texts to inquire about how central air conditioners are generally repaired and how to formulate maintenance plans;

[0061] The first maintenance model M2 extracts relevant information from the professional corpus T2 for the specific task of "central air-conditioning system maintenance", locates the search object of the specific problem, and inquires how to repair the specific model of central air-conditioning in this library; the other maintenance model M2 targets the specific task of "central air-conditioning system maintenance" and inquires how to formulate a maintenance plan for the specific model of central air-conditioning in this library.

[0062] Step S3.3, using the large model M1 and multiple small-scale neural network models M2 to collaborate, aggregate information and generate application responses:

[0063] After each small-scale neural network model M2 completes the query subtask, it converts the result into the conclusion in the structured format required by the large model M1 and sends it back to the large model M1;

[0064] The large model M1 receives the results processed by the small model M2, aggregates the conclusions of all small models, organizes them into friendly natural language, and provides query results for users.

[0065] Specifically, for example, after the two small models complete their subtasks, they convert the results into the structured format required by the large model M1. For example, the first model returns {"task":"Maintenance","device":"Central air conditioning","result":<Maintenance information>}, and the second model returns {"task":"Maintenance","device":"Central air conditioning","result":<Maintenance information>}; the large model M1 receives the conclusions processed by the small models, aggregates the conclusions of the two small models, organizes them into friendly natural language, and provides query results for managers, such as "The central air conditioning on the third floor of this library is a device from XX manufacturer, its maintenance process is..., and the current maintenance plan is...".

[0066] In summary, the present invention proposes a method of utilizing large language model technology to train a relatively large-scale language model and a series of smaller-scale professional models, which work together to provide delivery information to users and truly apply the useful information contained in the delivery materials.

[0067] The present invention intelligently processes building completion and delivery information through the collaborative work of a large model and multiple professional small models, thereby improving information processing efficiency, retrieval accuracy, and personalized service capabilities. The large model is responsible for understanding professional knowledge and complex information processing in the construction industry, while the small model focuses on efficient processing of specific tasks. The collaborative work of large and small models of the present invention can effectively handle a variety of complex data formats in the process of building completion and delivery, and improve the comprehensiveness and accuracy of information processing. Through the division of labor and cooperation of large and small models, the present invention can quickly extract the required information from massive data and flexibly apply it in different scenarios, greatly improving the efficiency of information retrieval.

[0068] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0069] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0070] Obviously, those skilled in the art can make various changes and modifications to the invention without departing from the spirit and scope of the invention. Thus, if these modifications and variations of the invention fall within the scope of the claims of the invention and their equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A digital delivery method for building completion with large and small models, characterized in that: include: Collect and organize a large amount of documents and data in the construction industry to form the basic corpus T1; Based on the special information of the completion and delivery stage of the current project, a professional corpus T2 for deep learning training is formed; Based on the constructed basic corpus T1, train the large model M1 in the vertical field; Train multiple small-scale neural network models M2 based on professional corpus T2; A large model M1 and multiple small-scale neural network models M2 are used to collaboratively collect and analyze basic data; The large model M1 and multiple small-scale neural network models M2 are used to collaborate and process the core tasks based on the delivered information; A large model M1 and multiple small-scale neural network models M2 are used to collaborate to aggregate information and generate application responses.

2. The method for digital delivery of building completion with large and small models as claimed in claim 1, characterized in that: Collect and organize a large amount of documents and data in the construction industry to form the basic corpus T1, including: Collect and organize a large amount of documents and data in the construction industry, perform text recognition, document tagging, document content association and analysis on architectural design drawings, construction records, equipment installation documents, and technical specifications, and form a basic corpus T1 through preprocessing.

3. The method for digital delivery of building completion with large and small models as claimed in claim 1, characterized in that: Based on the special information of the completion and delivery stage of the current project, a professional corpus T2 for deep learning training is formed, including: Automated annotation and keyword extraction are performed on the special information of the completion and delivery stage that is unique to the current project to form a professional corpus T2 for deep learning training; the professional corpus T2 is used to store key information of a single project in the process of building completion and delivery, providing a data basis for the training of the small-scale neural network model M2.

4. The method for digital delivery of building completion with large and small models as claimed in claim 1, characterized in that: Based on the constructed basic corpus T1, we train a large model M1 in the vertical field, including: The task of the big model M1 is to understand the professional knowledge of the construction industry, process documents in various formats, perform information extraction, document content analysis, knowledge question and answer, and code generation. The big model M1 is responsible for processing general tasks through a deep understanding of the semantics and context of the construction industry.

5. The method for digital delivery of building completion with large and small models as claimed in claim 1, characterized in that: Based on the professional corpus T2, multiple small-scale neural network models M2 are trained, including: Based on the professional corpus T2, multiple small-scale neural network models M2 are trained. Each small-scale neural network model M2 is used to perform a specific small task. Through the execution of each small task, specific problems in the completion and delivery of the building can be quickly processed, and the processed data can be converted into structured information. The structured information is written into the knowledge base for subsequent query and retrieval by the large model M1.

6. The method for digital delivery of building completion with large and small models as claimed in claim 5, characterized in that: The small tasks include: image recognition and restoration, CAD file content extraction, point cloud data processing, image detection and problem identification, path planning and task allocation.

7. The method for digital delivery of building completion with large and small models as claimed in claim 1, characterized in that: The large model M1 and multiple small-scale neural network models M2 are used to collaboratively collect and analyze basic data, including: Receive the user's natural language query through the large model M1 and understand the query intent; The large model M1 decomposes the query intent into specific query tasks and assigns them to the corresponding small-scale neural network model M2 according to the content of the query tasks; Based on the query task, the small-scale neural network model M2 further analyzes the query intent and generates several specialized query subtasks in combination with the specific content of this engineering project.

8. The method for digital delivery of building completion with large and small models as claimed in claim 1, characterized in that: The large model M1 and multiple small-scale neural network models M2 are used in collaboration to deliver information to process core tasks, including: The large model M1 searches for industry-level common texts or conducts large-scale searches and retrievals in massive amounts of delivery information documents based on query requirements; The small-scale neural network model M2 processes the core tasks of a specific type of data related to this project, focusing on the delivery information of this project, locating the search objects for specific problems, and completing query subtasks.

9. The method for digital delivery of building completion with large and small models as claimed in claim 1, characterized in that: The large model M1 and multiple small-scale neural network models M2 are used to collaborate to aggregate information and generate application responses, including: After each small-scale neural network model M2 completes the query subtask, it converts the result into the conclusion in the structured format required by the large model M1 and sends it back to the large model M1; The large model M1 receives the results processed by the small model M2, aggregates the conclusions of all small models, organizes them into friendly natural language, and provides query results for users.

Images