System and method for managing review data by adopting online large model

Abstract

The invention discloses a system and method for managing review data by adopting an online large model, and the method comprises the steps: building a data collection process, building an index algorithm, building a review opinion data large model, standardizing a data format, formulating a result standard, treating non-compliant data, checking key information, and achieving file archiving through an OSS service. Effective data support is provided for design review, and the method comprises the steps of basic data acquisition and integration, standardized formulation of a design review material list and accurate uploading and verification of design review materials; the method comprises the following steps: constructing a review opinion data large model, splitting historical review opinion data into large model components, supporting autonomous configuration of a review opinion model according to review specifications, finally splicing to generate a review opinion report, standardizing a data format, formulating an achievement standard, treating non-compliant data, and checking key information. And file archiving is realized through OSS service, and effective data support is provided for design review.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a management system and method for reviewing data using an online large model. Background Art

[0002] In accordance with the requirements of the "Administrative Measures for the Approval of Preliminary Design of Power Transmission and Transformation Projects of State Grid Corporation of China" (State Enterprise Management

[2019] No. 426) and the "Regulations on the Depth of Preliminary Design Content of Power Transmission and Transformation Projects of State Grid Corporation of China", before the review plan is arranged, the project legal person unit organizes an internal review of the preliminary design documents submitted by the design unit to form an internal review opinion; 7 working days before the review meeting, the preliminary design documents shall be submitted to the review unit; before the review meeting, the review unit shall conduct a preliminary review of the preliminary design documents and form a preliminary review opinion; 3 working days after the preliminary design is approved, the design unit shall submit the approved design documents and budget estimates; after the opinions are issued, the assistant project manager shall collect the design results, meeting notices, review minutes sign-in sheets and other materials in accordance with the requirements for archiving the materials.

[0003] At present, review materials are still managed offline and lack effective information technology to collect them. The following problems exist: First, due to the large number of design units, the submitted materials are not timely and complete, lack unified standards, and online management has not been realized; second, the review-related technical indicators still need to be manually searched, and structured extraction has not been realized; third, the review materials are not collected in time during the process, resulting in a lot of time required for collection in the later stage, inconsistent versions, and a large workload for data collection.

[0004] Therefore, it is urgently necessary to make full use of digital technology, focus on the core of "review business", accelerate the evolution of existing systems, improve the six core business capabilities of full-process management and control, data collection, mobile Internet, opinion collaboration, resource sharing, and decision-making analysis, realize the four digitalizations of "business, data, management, and collaboration", create a design review management system, and build a new review model driven by "digital technology + data elements", promote the digital upgrade of the entire process, all elements, and all participants of the review business, and provide strong support for the implementation of the construction goals of the Digital Economic Research Institute. Summary of the invention

[0005] The purpose of the present invention is to solve at least one of the technical problems existing in the prior art, and to provide a management system and method for review materials using an online big model. By constructing a data collection module, it assists in the rapid archiving of actual project design review work materials, and by constructing a key cost indicator algorithm, it builds an OCR indicator recognition technology to quickly generate review indicators, realize structured data extraction of table indicators, and realize unstructured data extraction of full-text indicators and paragraph indicators. The extracted indicators are automatically reviewed according to the design review specifications. By constructing a big model of review opinion data, the historical review opinion data is split into big model components, and support is provided for the independent configuration of the review opinion model according to the review specifications. Finally, the review opinion report is generated by splicing, and the data format is standardized, the achievement standard is formulated, the non-compliant data is governed, the key information is checked and verified, and the file is archived through the OSS service, so as to provide effective data support for the design review.

[0006] The present invention also provides a method for managing review materials using an online big model, including establishing a data collection process, establishing an indicator algorithm, and constructing a review opinion data big model. The method is characterized in that the data collection process is established by standardizing data formats, formulating achievement standards, managing non-compliant data, checking and verifying key information, and archiving files through OSS services to provide effective data support for design review, including: basic data acquisition and integration, standardized formulation of a design review material list, and accurate uploading and verification of design review materials;

[0007] The index establishment algorithm extracts relevant indicators from the original data by applying the data sorting results, cleans the data, removes redundant information, sets keyword extraction, and uses natural language processing technology to identify keywords related to the indicators;

[0008] The construction review opinion data model is based on the collection and sampling of national network design review big data, extracts relevant review opinions, uses natural language processing technology to extract key information and customizes review standards and requirements for various types of projects, modularizes the review opinions, combines them according to different project types, configures audit points, suggestions and precautions, and finally automatically generates a complete review report.

[0009] According to a method for managing review materials using an online large model provided by the present invention, the establishment of an indicator algorithm includes the following steps:

[0010] The first step is to record the keywords of the design specifications corresponding to the indicators extracted in the actual work scenarios of the review business through the annotation platform, store their location information in a structured manner, and build a business sample library;

[0011] The second step is to split the samples of different indicator models into training sets and test sets, configure the recognition algorithms respectively, write machine learning models based on regular expressions, and recognize indicator formats;

[0012] The third step is to use Docker to build a model training environment on the paddle NLP platform, copy and mount the training and debugging Paddle NLP source code to the specified directory, enter the container to install the underlying dependencies of the training model, and finally implement joint debugging through business code. Write the corresponding business implementation code and the client httpc l ient calling code that calls the deployment service to perform joint debugging, realize indicator visualization, and informationize the review.

[0013] According to a method for managing review materials using an online large model provided by the present invention, the basic data is acquired and integrated, and project information is synchronized to the system through an external platform. The system comprehensively collects and reserves project-related information, including core elements such as project code, project type, and project scale. The project review schedule is acquired synchronously through the system, and key information matching technology is used to automatically generate a project milestone plan schedule, providing project personnel and review experts with a reasonable allocation of project design review related work. A comprehensive review resource library is built, which deeply integrates multiple data resources such as the State Grid Information Library, the General Cost Library, the multi-dimensional reference price, and the document database, providing data support and guarantee for the generation of review indicators.

[0014] According to a method for managing review materials using an online big model provided by the present invention, the standardized formulation of the design review material list, through the design review material list, formulates material lists and data format specifications for different project types, unifies design achievement standards, and reduces the differences in achievements between similar design units. The consistency and standardization of design achievement management are improved, laying a solid foundation for subsequent review work.

[0015] According to a method for managing review materials using an online large model provided by the present invention, the design review materials are accurately uploaded and verified, including design specifications, design drawings, engineering quantity calculations, equipment and material lists, etc. Through the automatic verification function, the uploaded materials are reviewed twice to further ensure the smooth progress of the review work.

[0016] According to a method for managing review materials using an online big model provided by the present invention, the construction of the review opinion data big model includes: setting up a review model, generating design review opinions based on the AI ​​big model, setting up the review model according to the form and method of the review, by summarizing the review content, key points and specifications, formulating unified standards, through refined classification, performing corresponding configuration according to the classification, and through classification management, subdividing the model into a table model, a paragraph model and a full-text model.

[0017] According to a method for managing review materials using an online big model provided by the present invention, the design review opinions are generated based on the AI ​​big model, and detailed design review indicators are formulated according to the pricing methods and standards for power engineering cost design. Based on this, an efficient and accurate model algorithm is constructed, and pre-training and expansion are performed according to standard pricing methods and standards. By creating an indicator enhancement framework, the processing capability of logical relationships of large quantities of documents is improved, multi-task fine-tuning is implemented, and the review opinion task is combined with other related tasks such as professional classification of review projects, professional classification of expert review, etc. The generalization ability of the model is enhanced through joint training, and the design review opinion big data resources of provinces across the country are used to construct a structured review model and a model library for configuring unstructured components.

[0018] According to the present invention, a management system for review materials using an online large model includes: a user layer, an application layer, a service layer, a data layer and an infrastructure layer. The user layer accesses the system through a PC application and an APP mobile terminal. The application layer includes multiple functional modules such as plan management, opinion management and review meeting management, which are responsible for realizing various business functions of the system.

[0019] The service layer includes authority management services, data storage services, and file storage service interfaces; the data layer includes project information, review plans, review opinion data, review resource data, and unstructured data; the infrastructure layer provides hardware and network basic support, including servers and network equipment, to ensure the stable operation of the system.

[0020] Compared with the prior art, the present invention adopts an online large model to manage and manage review materials, and by constructing a data collection module, assists in the rapid archiving of actual project design review work materials.

[0021] Compared with the prior art, the present invention adopts an online large model to manage and manage review materials. By constructing a key cost indicator algorithm and setting up OCR indicator recognition technology, review indicators are quickly generated, structured data extraction of table indicators is realized, and unstructured data extraction of full-text indicators and paragraph indicators is realized, and the extracted indicators are automatically reviewed according to design review specifications.

[0022] Compared with the prior art, the present invention adopts an online big model to manage and manage review materials. By building a big model of review opinion data, the historical review opinion data is split into big model components, supporting the independent configuration of the review opinion model according to the review specification, and finally splicing and generating the review opinion report. By standardizing the data format, formulating the achievement standard, managing the non-compliant data, checking and verifying the key information, and realizing file archiving through the OSS service, effective data support is provided for the design review. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The present invention is further described below in conjunction with the accompanying drawings and embodiments;

[0024] Figure 1 This is a structural diagram of a management system for reviewing materials using an online large model according to the present invention;

[0025] Figure 2 The present invention is a flowchart of a method for managing review materials using an online large model. DETAILED DESCRIPTION

[0026] This section will describe in detail the specific embodiments of the present invention. The preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present invention, but it cannot be understood as a limitation on the scope of protection of the present invention.

[0027] Reference Figure 1 The embodiment of the present invention adopts an online big model to manage the review materials, including a user layer, an application layer, a service layer, a data layer and an infrastructure layer. The user layer accesses the system through PC applications and APP mobile terminals. The application layer includes multiple functional modules such as plan management, opinion management, and review meeting management, and is responsible for realizing various business functions of the system; the service layer includes permission management services, data storage services, and file storage service interfaces; the data layer includes project information, review plans, review opinion data, review resource data, and unstructured data; the infrastructure layer provides hardware and network basic support, including servers and network equipment, to ensure the stable operation of the system.

[0028] Reference Figure 2 The embodiment of the present invention is a method for managing review materials using an online big model, including establishing a data collection process, establishing an indicator algorithm, and building a review opinion data big model. The specific implementation steps of establishing the data collection process are as follows:

[0029] Step 1: Acquisition and integration of basic data;

[0030] To ensure the comprehensiveness and accuracy of project review, the system first synchronizes project information to the system through an external platform, and the system comprehensively collects and stores project-related information, including core elements such as project code, project type, and project scale. At the same time, the system simultaneously obtains the project review schedule, and uses key information matching technology to automatically generate a project milestone schedule to provide project personnel and review experts with a reasonable allocation of project design review-related work.

[0031] In addition, in order to improve the accuracy and efficiency of the review work, a comprehensive review resource library has been built. This resource library deeply integrates multiple data resources such as the State Grid Information Database, the General Cost Database, the Multi-dimensional Reference Price, and the Document Database, providing data support and guarantee for the generation of review indicators.

[0032] Step 2: Standardize the design review materials list;

[0033] In view of the characteristics of different project types, differentiated design review material lists were designed, and strict data format specifications and unified standards for design results were formulated. This measure aims to control data quality from the source and promote standardized management of design results. By reducing the differences in results between similar design units, the consistency and standardization of design results management are improved, laying a solid foundation for subsequent review work.

[0034] Step 3: Accurate upload and verification of design review materials;

[0035] According to the requirements of the established project type list, upload the relevant information of the design review materials, including design specifications, design drawings, engineering quantity calculations, equipment and material lists and other key contents. During the upload process, strictly follow the standard process to ensure the integrity and accuracy of the uploaded materials. At the same time, the system also has an automatic verification function to conduct a secondary review of the uploaded materials to further ensure the smooth progress of the review work.

[0036] The specific implementation steps for building a large model of review opinion data are as follows:

[0037] Step 1: Set up the review model;

[0038] Based on the forms and methods of design review in previous years, we systematically summarized and analyzed the review content, key points and specifications, and formulated a unified standard design review model based on this. This model supports free configuration and resource sharing, aiming to improve review efficiency and quality.

[0039] For different types of projects, such as substations, lines, cables, etc., we have carried out detailed classification and configured corresponding design review models. Through this measure, we can further refine the content of the review results, thus ensuring the steady improvement of the review quality.

[0040] In the management of the large review model, a classification management approach is adopted, which subdivides the model into table model, paragraph model and full-text model.

[0041] The table model focuses on the form information in the review materials. Through recognition and extraction technology, combined with model algorithms, it automatically calculates the review indicators to ensure the accuracy and completeness of the data.

[0042] The paragraph model focuses on the paragraph content in the review materials. Through keyword recognition and extraction technology, it is effectively applied to the generation of review opinion results, thereby improving the pertinence and accuracy of the review opinions.

[0043] The full-text model uses model algorithms to identify and extract key information from non-structured and difficult-to-accurately locate paragraphs. The model has been rigorously trained to ensure the accuracy of extraction, helping reviewers to quickly locate key information, thereby speeding up the secondary revision process of review opinions and making design review work more efficient and accurate.

[0044] Step 2: Generate design review opinions based on the AI ​​big model;

[0045] According to the "Transmission Line Engineering Quantity Pricing Specifications", "Substation Engineering Quantity List Pricing Specifications", "Construction Engineering Quantity List Pricing Specifications" and "Power Construction Project Budget Quotas" and other specifications, we strictly follow the pricing methods and standards for power engineering cost design, formulate detailed design review indicators, and build an efficient and accurate model algorithm based on this.

[0046] In the preprocessing stage, in order to improve the adaptability and accuracy of the model, pre-training expansion was carried out. First, a large number of historical review opinions were analyzed by big data, and the characteristic values ​​of target opinions closely related to various review types were carefully sorted out and extracted, and a clear correspondence was established. Subsequently, these valuable review opinion corpora were fully utilized to conduct in-depth training of the model, so that the model can accurately target the current review project and efficiently perform the task of extracting key indicators of review opinions.

[0047] In order to further explore the potential of the model, an indicator enhancement framework was innovatively designed based on the review opinion LLM model. The introduction of this framework greatly improved the ability to handle large quantities of document logical relationships, allowing the model to maintain a high degree of accuracy and stability in complex and changing review scenarios. By using the extracted indicators and complete document content, the model can accurately predict the entities corresponding to the review opinions, thereby ensuring that the review opinions that meet the requirements of the specifications and are close to the actual situation are finally generated.

[0048] During the model fine-tuning phase, a variety of strategies were adopted to further improve the accuracy of review opinion generation. First, multi-task fine-tuning was implemented to combine the review opinion task with other related tasks such as review engineering professional classification and expert review professional classification, and enhance the generalization ability of the model through joint training. At the same time, the performance of the model was closely monitored, and the training parameters or optimization algorithms were adjusted in a timely manner according to the evaluation results to ensure that the model was always in the best state. In addition, a manual intervention mechanism was introduced to use the rich experience and expertise of experts to fine-tune and optimize the model output to reduce the occurrence of erroneous review opinions and improve the quality of review opinions.

[0049] Finally, we made full use of the big data resources of design review opinions from provinces across the country to build a structured review model and a model library for configuring unstructured components. The generation of structured general models for review opinions and the free configuration of unstructured component models have greatly improved the efficiency and quality of design review work. After automatically generating a general model according to the project type, we can flexibly select component models and quickly generate design review opinion results that meet the requirements, providing strong support for power engineering construction.

[0050] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the technical field without departing from the purpose of the present invention.

Claims

1. A method for managing review materials using an online big model, including establishing a data collection process, establishing an indicator algorithm, and constructing a review opinion data big model, characterized in that: The process of establishing data collection is to standardize data formats, formulate achievement standards, manage non-compliant data, check and verify key information, and archive files through OSS services to provide effective data support for design review, including: basic data acquisition and integration, standardized formulation of design review material lists, and accurate uploading and verification of design review materials; The index establishment algorithm extracts relevant indicators from the original data by applying the data sorting results, cleans the data, removes redundant information, sets keyword extraction, and uses natural language processing technology to identify keywords related to the indicators; The construction review opinion data model is based on the collection and sampling of national network design review big data, extracts relevant review opinions, uses natural language processing technology to extract key information and customizes review standards and requirements for various types of projects, modularizes the review opinions, combines them according to different project types, configures audit points, suggestions and precautions, and finally automatically generates a complete review report.

2. According to claim 1, a method for managing review materials using an online big model is characterized in that: The index establishment algorithm comprises the following steps: The first step is to record the keywords of the design specifications corresponding to the indicators extracted in the actual work scenarios of the review business through the annotation platform, store their location information in a structured manner, and build a business sample library; The second step is to split the samples of different indicator models into training sets and test sets, configure the recognition algorithms respectively, write machine learning models based on regular expressions, and recognize indicator formats; The third step is to use Docker to build a model training environment on the PaddlePaddle platform, copy and mount the PaddleNLP source code for training and debugging to the specified directory, enter the container to install the underlying dependencies of the training model, and finally implement joint debugging through business code. Write the corresponding business implementation code and the client httpclient calling code that calls the deployment service to perform joint debugging, realize indicator visualization, and information-based review.

3. The method for managing review materials using an online big model according to claim 1, characterized in that: The basic data acquisition and integration described above synchronizes project information to the system through an external platform. The system comprehensively collects and reserves project-related information, including core elements such as project code, project type, and project scale. The system synchronously obtains the project review plan, and uses key information matching technology to automatically generate a project milestone plan schedule, providing project personnel and review experts with a reasonable allocation of project design review related work. A comprehensive review resource library has been built, which deeply integrates multiple data resources such as the State Grid Information Library, the General Cost Library, the multi-dimensional reference price, and the document database.

4. The method for managing review materials using an online big model according to claim 1, characterized in that: The standardized formulation of the design review materials list is used to formulate material lists and data format specifications for different project types and unify design results standards.

5. The method for managing review materials using an online big model according to claim 1, characterized in that: The accurate uploading and verification of the design review materials, including design specifications, design drawings, quantity calculations, and equipment and material lists, conducts a secondary review of the uploaded materials through the automatic verification function.

6. The method for managing review materials using an online big model according to claim 1, characterized in that: The construction of the review opinion data big model includes: setting up a review model and generating design review opinions based on the AI ​​big model. The setting up of the review model is based on the form and method of the review, by summarizing the review content, key points and specifications, formulating unified standards, and through refined classification, performing corresponding configuration according to the classification. Through classification management, the model is subdivided into table model, paragraph model and full-text model.

7. A method for managing review materials using an online big model according to claim 6, characterized in that: The design review opinions are generated based on the AI ​​big model, and detailed design review indicators are formulated according to the pricing methods and standards of power engineering cost design. Based on this, an efficient and accurate model algorithm is constructed, and pre-training and expansion are carried out according to standard pricing methods and standards. By creating an indicator enhancement framework, the processing capability of logical relationships of large quantities of documents is improved, and multi-task fine-tuning is implemented. The review opinion task is combined with other related tasks such as review engineering professional classification, expert review professional classification, etc. The generalization ability of the model is enhanced through joint training, and the design review opinion big data resources of provinces across the country are used to construct a structured review model and a model library for configuring unstructured components.

8. A management system for review materials using an online large model, characterized by A method for managing review materials using an online large model as described in any one of claims 1 to 7 is adopted, comprising: a user layer, an application layer, a service layer, a data layer and an infrastructure layer, wherein the user layer accesses the system through a PC application and an APP mobile terminal, and the application layer includes multiple functional modules such as plan management, opinion management and review meeting management, which are responsible for realizing various business functions of the system; The service layer includes authority management services, data storage services, and file storage service interfaces; the data layer includes project information, review plans, review opinion data, review resource data, and unstructured data; the infrastructure layer provides hardware and network basic support, including servers and network equipment, to ensure the stable operation of the system.

Images