A large model-based power engineering supervision business intelligent processing method and system

Abstract

The application discloses a large model-based power engineering supervision business intelligent processing method and system, and relates to the technical field of engineering supervision intelligentization. The method obtains a target supervision business request and related multi-source supervision business data, performs format analysis and semantic extraction on the multi-source supervision business data to obtain a business object, an object correlation and a source identifier; constructs a supervision business knowledge index including at least one of a vector index, a keyword index and a structured correlation index based on the same; identifies a business processing type and a processing strategy corresponding to the request, retrieves a target business object, a target object correlation and a target source identifier to generate a business context; calls a corresponding prompt word template and / or a business processing rule to input a large model to generate an initial processing result, and outputs a target processing result after consistency checking and traceability marking. Thus, the supervision business data utilization efficiency is improved, the manual dependence for data processing and examination is reduced, and the risk information comprehensive identification capability is enhanced.

Description

Technical Field

[0001] This application relates to the field of intelligent engineering supervision technology, and in particular to an intelligent processing method and system for power engineering supervision business based on a large model. Background Technology

[0002] With the continuous increase in the scale and management requirements of power engineering construction, the project data, site records, personnel information, quality and safety data, and management processes involved in engineering supervision are becoming increasingly complex. Supervision units need to continuously track and monitor construction progress, quality control, safety management, document compliance, and on-site personnel operations during the construction process. Related business data typically originates from different business stages and information systems, and the data types cover various formats such as text, tables, images, approval records, and personnel qualification information. Therefore, how to improve the efficiency of information acquisition, data processing, and risk discovery in engineering supervision has become a crucial issue in the informatization of engineering supervision.

[0003] Currently, engineering supervision typically utilizes information technology tools such as project management systems, mobile data entry tools, and electronic document management systems to manage project data, process approvals, on-site records, and personnel information. While these methods have achieved a certain degree of electronic storage and streamlined workflow of business data, in practice, the large amount of business data accumulated in the systems often exists in a scattered and static manner, with weak correlation between different documents. When supervisors search for professional documents, verify business data, compile supervision documents, or review construction-related materials, they still need to rely on manual retrieval, comparison, and experience-based judgment, resulting in low data utilization efficiency, lengthy processing times, and potential discrepancies in judgment standards among different personnel regarding similar issues. Furthermore, existing methods often struggle to comprehensively identify issues such as on-site personnel qualifications, operational risks, and missing data from multiple sources in a timely manner, affecting the standardization of supervision work and the effectiveness of risk control.

[0004] Therefore, in the informatization of power engineering supervision, the problems that urgently need to be solved include the fragmentation of business data and low utilization efficiency, the high degree of manual reliance on supervision data processing and review, and the insufficient comprehensive identification capability of on-site risk information. Summary of the Invention

[0005] This application provides an intelligent processing method and system for power engineering supervision business based on a large model, aiming to solve the problems of scattered business data and low utilization efficiency, high reliance on manual labor in the information processing of power engineering supervision business, and insufficient comprehensive identification capability of on-site risk information in the existing technology.

[0006] Firstly, a method for intelligent processing of power engineering supervision business based on a large model, the method comprising: Obtain the target supervision business request and multi-source supervision business data related to the target supervision business request. The multi-source supervision business data includes at least one of project data, on-site record data, personnel information data, quality and safety data, and management process data. The multi-source supervision business data is parsed and semantically extracted to obtain business objects, object relationships between business objects, and source identifiers used to characterize the source of the business objects; Based on the business object, the object association, and the source identifier, a supervision business knowledge index is constructed, which includes at least one of vector index, keyword index, and structured association index. The intent of the target supervision business request is identified to determine the business processing type and processing strategy corresponding to the target supervision business request. The business processing type includes at least one of information query, data analysis, document generation, document review and risk verification. Based on the business processing type and the processing strategy, retrieve the target business object, target object association, and target source identifier that match the target supervision business request from the supervision business knowledge index, and generate a business context based on the target business object, the target object association, and the target source identifier; Based on the business processing type, the corresponding prompt word template and / or business processing rule are invoked, the target supervision business request and the business context are input into the large model, and the initial processing result is generated. Based on the processing strategy, the target object association, and the target source identifier, the initial processing result is subjected to consistency verification and source tracing annotation to obtain the target processing result; Output the target processing result.

[0007] Optionally, in the above scheme, the step of performing format parsing and semantic extraction on the multi-source supervision business data to obtain business objects, object relationships between the business objects, and source identifiers used to characterize the source of the business objects includes: Identify the data format and data source of the multi-source supervision business data to obtain data format identifier and data source identifier; Based on the data format identifier, the corresponding parsing strategy is invoked to parse the multi-source supervision business data to obtain standardized business content that includes at least one of the following: text content, table content, image content, title content, paragraph content, header and footer content, and process record content. Based on the layout, content level, and contextual semantics of the standardized business content, the standardized business content is reorganized in order to obtain business content to be extracted that conforms to the business reading order. Entity recognition and attribute extraction are performed on the business content to be extracted to obtain the business object and the object attributes corresponding to the business object. Based on the matching relationship between the context position of the business object in the business content to be extracted and the object attribute, the object association relationship between the business objects is determined, and the data source identifier is bound to the business object to obtain the source identifier.

[0008] Optionally, in the above scheme, constructing a supervision business knowledge index based on the business object, the object association, and the source identifier includes: Based on the object attributes of the business object, the business object is classified into business categories to obtain at least one of the following categories: project object, document object, personnel object, quality and safety object, process object, and standard basis object. The classification objects and the standardized business content corresponding to the classification objects are vectorized to obtain vector indexes; Keyword index is obtained by extracting keywords and generating tags from the classification objects and their corresponding object attributes. Based on the object association relationships, a structured association index is generated to represent the business association paths between different categories of objects; The source identifier is mapped to the vector index, the keyword index, and the structured association index respectively to obtain the supervision business knowledge index.

[0009] Optionally, in the above scheme, the step of performing intent recognition on the target supervision business request and determining the business processing type and processing strategy corresponding to the target supervision business request includes: The target supervision business request is semantically parsed to obtain the request action, request object, query conditions and result requirements; Based on the request action and the request object, determine the business processing type corresponding to the target supervision business request; Based on the query conditions and the result requirements, determine the retrieval scope, retrieval method, output format, and verification rules corresponding to the business processing type; The processing strategy is generated based on the business processing type, the search scope, the search method, the output format, and the verification rules.

[0010] Optionally, in the above scheme, when the business processing type is an information query type, the step of retrieving the target business object, target object association, and target source identifier matching the target supervision business request from the supervision business knowledge index according to the business processing type and the processing strategy, and generating a business context based on the target business object, the target object association, and the target source identifier, includes: Based on the retrieval scope and retrieval method in the processing strategy, candidate business objects and candidate object associations are retrieved from at least one of the vector index, the keyword index, and the structured association index. Based on the semantic relevance, keyword matching degree, and association path matching degree between the candidate business objects and the target supervision business request, the candidate business objects and their association relationships are sorted to obtain the target business objects and their association relationships. The target source identifier is determined based on the source identifier corresponding to the target business object and the association relationship of the target object; Based on the target business object, the target object's association relationship, and the target source identifier, a question-and-answer context is generated to answer the target supervision business request; The target supervision business request and the question-and-answer context are input into the large model to obtain the question-and-answer results, and the question-and-answer results are used as the initial processing results.

[0011] In the above scheme, optionally, when the business processing type is data analysis, the step of calling the corresponding prompt word template and / or business processing rule according to the business processing type, inputting the target supervision business request and the business context into the large model, and generating the initial processing result includes: Based on the target supervision business request, determine the business object to be analyzed, the analysis indicators, the analysis dimensions, the screening conditions, and the result display requirements; Based on the business object to be analyzed, the analysis indicators, the analysis dimensions, and the filtering conditions, the target data table, target fields, and field relationships are determined from the structured association index. Based on the target data table, the target fields, and the relationships between the fields, a business data query statement is generated; The business data query statement is subjected to syntax validation, field permission validation, and query range validation to obtain a valid business data query statement. If the business data query statement fails the validation, the business data query statement is corrected according to the reason for the validation failure, resulting in a corrected business data query statement. Execute the verified business data query statement or the corrected business data query statement to obtain the business query result; Based on the business query results and the result display requirements, at least one of the following is generated: data analysis text, chart configuration information, and data export information. At least one of the following is used as the initial processing result: data analysis text, chart configuration information, and data export information.

[0012] Optionally, in the above scheme, when the business processing type is document generation, the step of calling the corresponding prompt word template and / or business processing rule according to the business processing type, inputting the target supervision business request and the business context into the large model, and generating the initial processing result includes: Based on the target supervision service request, determine the document type and purpose of the document to be generated; Based on the document type and the document purpose, obtain the corresponding power engineering supervision document template to obtain the target document template; Based on the field items and chapter items in the target document template, the corresponding project fact information, process record information, quality and safety information, and standard basis content are extracted from the business context to obtain template filling information; Establish a correspondence between the source identifier corresponding to the template fill information and the field items and chapter items in the target document template to obtain the field source mapping relationship; Based on the field source mapping relationship, the template filling information is filled into the target document template to obtain the document draft content; Based on the format requirements, content integrity requirements, and field source mapping relationships corresponding to the target document template, the content of the document draft is completed and standardized to obtain the content of the supervision document, and the content of the supervision document is used as the initial processing result.

[0013] In the above scheme, optionally, when the business processing type is document review, the step of calling the corresponding prompt word template and / or business processing rule according to the business processing type, inputting the target supervision business request and the business context into the large model, and generating the initial processing result includes: The construction documents to be reviewed are determined from the target supervision service request or the multi-source supervision service data; The construction documents to be reviewed are subjected to chapter identification and content element extraction to obtain the structured content of the construction documents. Based on the structured content of the construction data, determine the compliance review items, completeness review items, matching degree review items, and risk logic review items; Based on the compliance review items, the integrity review items, the matching degree review items, and the risk logic review items, the corresponding review basis is retrieved from the supervision business knowledge index to obtain the target review basis; Based on the aforementioned target review criteria, the structured content of the construction data is subjected to standard clause matching, key module missing detection, construction content consistency comparison, and risk logic verification to obtain problem identification results. Based on the problem identification results and the target review criteria, a structured review result containing problem points, supporting content, and rectification suggestions is generated, and the structured review result is used as the initial processing result.

[0014] Optionally, in the above scheme, when the business processing type is risk verification, the step of retrieving the target business object, target object association, and target source identifier matching the target supervision business request from the supervision business knowledge index according to the business processing type and the processing strategy, and generating a business context based on the target business object, the target object association, and the target source identifier, includes: Acquire on-site identification data collected by on-site acquisition equipment; The on-site identification data is used to identify objects to obtain on-site objects to be verified, including personnel to be verified. The individuals to be verified are matched with the baseline information of the individuals to obtain the identity matching result and the identity matching confidence level. If the identity matching confidence level meets the preset identity confirmation conditions, the corresponding personnel information, entry approval information, qualification certificate information, examination record information and violation record information are retrieved from the supervision business knowledge index according to the identity matching result to obtain the on-site object association data; Based on the on-site object association data, determine the identity status, application status, qualification status, examination status, and historical risk status of the personnel to be verified; The identity status, the application status, the qualification status, the examination status, and the historical risk status are matched with preset risk verification rules to obtain the risk hit result; Based on the risk hit results, a risk verification result is generated, and the risk verification result is used as the initial processing result.

[0015] Secondly, a large-scale model-based intelligent processing system for power engineering supervision business, the system comprising: The data acquisition module is used to acquire the target supervision business request and multi-source supervision business data related to the target supervision business request. The multi-source supervision business data includes at least one of project data, on-site record data, personnel information data, quality and safety data, and management process data. The parsing and extraction module is used to perform format parsing and semantic extraction on the multi-source supervision business data to obtain business objects, object association relationships between the business objects, and source identifiers used to characterize the source of the business objects; An index building module is used to build a supervision business knowledge index based on the business object, the object association relationship, and the source identifier. The supervision business knowledge index includes at least one of vector index, keyword index, and structured association index. The intent recognition module is used to recognize the intent of the target supervision business request and determine the business processing type and processing strategy corresponding to the target supervision business request. The business processing type includes at least one of information query, data analysis, document generation, document review and risk verification. The context generation module is used to retrieve, according to the business processing type and the processing strategy, the target business object, the target object association relationship and the target source identifier that match the target supervision business request from the supervision business knowledge index, and generate a business context based on the target business object, the target object association relationship and the target source identifier; The large model processing module is used to call the corresponding prompt word template and / or business processing rules according to the business processing type, input the target supervision business request and the business context into the large model, and generate the initial processing result; The verification and tracing module is used to perform consistency verification and tracing annotation on the initial processing result based on the processing strategy, the target object association relationship and the target source identifier, so as to obtain the target processing result; The result output module is used to output the target processing result.

[0016] Compared with the prior art, this application has at least the following beneficial effects: This application, based on further analysis and research of existing technical problems, recognizes the following issues in the information processing of power engineering supervision: scattered business data with low utilization efficiency, high reliance on manual labor in supervision data processing and review, and insufficient comprehensive identification capabilities for on-site risk information. By acquiring target supervision business requests and related multi-source supervision business data, and performing format parsing and semantic extraction on this multi-source data, the content originally scattered across project documents, on-site records, personnel information, quality and safety data, and management process data is transformed into business objects, object relationships, and source identifiers. This provides a foundation for unified organization and related utilization of supervision business data from different sources and in different formats. Furthermore, based on business objects, object relationships, and source identifiers, a supervision business... The knowledge index enables the retrieval of supervision business data based on semantics, keywords, and business-related paths, improving the utilization efficiency of scattered business data. Building upon this, intent recognition of target supervision business requests determines the business processing type and strategy. Based on this, matching target business objects, target object relationships, and target source identifiers are retrieved to generate business context. This allows the large model's processing to be based on business content relevant to the current request and with provenance, rather than relying on manual item-by-item searching and experience-based judgment. Simultaneously, based on the business processing type, corresponding prompt word templates and / or business processing rules are invoked to generate initial processing results. These results are then validated for consistency and traceability based on the processing strategy, target object relationships, and target source identifiers. This ensures that the output target processing results balance automatic generation efficiency, business consistency, and traceability. Therefore, this application can improve the efficiency of multi-source data integration and utilization in power engineering supervision, reduce the reliance on manual experience in supervision data processing and review, and enhance the comprehensive identification and processing capabilities of risk information related to supervision, thereby solving the problems of scattered business data and low utilization efficiency, high reliance on manual labor in supervision data processing and review, and insufficient comprehensive identification capabilities of on-site risk information. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an embodiment of the intelligent processing method for power engineering supervision business based on a large model provided in this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] In one embodiment, such as Figure 1As shown, a method for intelligent processing of power engineering supervision business based on a large model is provided. This method can be applied to a power engineering supervision business management platform, a mobile application for supervision, or a business system formed by combining the two. This business system can adopt a front-end and back-end separated system architecture, allowing users to initiate target supervision business requests through personal computers, mobile terminals, or interface servers. Target supervision business requests can be text requests, text requests converted from speech, form-triggered requests, or requests automatically generated by business process nodes, such as engineering data query requests, project data analysis requests, supervision document generation requests, construction data review requests, and on-site personnel risk verification requests.

[0020] When performing intelligent processing, the first step is to acquire the target supervision business request and related multi-source supervision business data. This multi-source supervision business data can originate from project data management systems, on-site recording systems, personnel information systems, quality and safety management systems, process approval systems, document management systems, knowledge base management systems, and on-site data acquisition equipment access systems, etc. Multi-source supervision business data can include at least one of the following: project data, on-site recording data, personnel information data, quality and safety data, and management process data. Specifically, project data can include construction plans, supervision plans, supervision details, monthly supervision reports, supervision logs, acceptance data, and project progress data; on-site recording data can include on-site observation records, inspection records, on-site photos, video screenshots, and data collected by equipment recorders; personnel information data can include personnel identity information, entry approval information, safety regulation examination records, special operation qualification information, and violation records; quality and safety data can include quality problems, safety hazards, rectification closure records, and risk event records; and management process data can include approval records, workflow status, operation logs, and processing opinions.

[0021] After acquiring multi-source supervision business data, the data undergoes format parsing and semantic extraction to obtain business objects, object relationships between business objects, and source identifiers representing the origin of business objects. Specifically, the corresponding parsing strategy is first invoked according to the data format to convert data of different formats into unified standardized business content. Then, business objects are identified through entity recognition, attribute extraction, relationship extraction, and context analysis. Business objects can include engineering projects, construction units, supervision personnel, operators, construction documents, specification clauses, approval items, quality issues, safety hazards, and risk records. Object relationships can include relationships between projects and documents, personnel and qualifications, construction content and specification basis, quality and safety issues and rectification records, and approval items and process status. Source identifiers can be used to represent information such as the source system, source document, document page number, paragraph position, table cell, process node, acquisition device, and acquisition time of the business object.

[0022] After obtaining the business objects, their relationships, and source identifiers, a supervision business knowledge index is constructed based on these elements. This index can include at least one of vector indexes, keyword indexes, and structured relational indexes. Vector indexes support semantic retrieval, keyword indexes support exact matching and tag filtering, and structured relational indexes support tracking relationships and path retrieval between business objects. When constructing the supervision business knowledge index, text fragments, table content, specification clauses, process records, etc., corresponding to the business objects can be converted into vector representations. Keyword indexes are then applied to the names, attributes, tags, and business categories of the business objects. Simultaneously, a structured relational index is constructed based on the object relationships, enabling unified retrieval and association of data from different sources, in different formats, and from different business stages.

[0023] Upon receiving a target supervision business request, the intent of the request is identified to determine the corresponding business processing type and strategy. Business processing types can include at least one of the following: information query, data analysis, document generation, document review, and risk verification. Processing strategies can include search scope, search methods, prompt word templates, business processing rules, output format, permission requirements, verification rules, and traceability requirements. For example, for information query requests, the strategy might require prioritizing the retrieval of question-and-answer pairs, regulatory clauses, and project materials in the knowledge base; for data analysis requests, the strategy might require determining the data tables, fields, analysis dimensions, and chart display methods; for document review requests, the strategy might require processing based on dimensions such as compliance, completeness, matching degree, and risk logic; and for risk verification requests, the strategy might require a comprehensive judgment based on identity status, approval status, qualification status, examination status, and historical risk status.

[0024] After determining the business processing type and strategy, the system retrieves the target business object, its associated relationships, and the target source identifier that match the target supervision business request from the supervision business knowledge index. A business context is then generated based on these elements. The business context can include factual content, regulatory basis, business data, historical processing records, source identifiers, associated paths, and necessary output constraints related to the target supervision business request. This business context allows subsequent large-scale model processing to be executed based on searchable, associative, and traceable business content, rather than solely relying on general language generation capabilities.

[0025] After generating the business context, the corresponding prompt word template and / or business processing rules are invoked based on the business processing type. The target supervision business request and business context are input into the large model to generate initial processing results. Prompt word templates can be configured according to business categories, such as knowledge Q&A prompt word templates, data analysis prompt word templates, supervision document generation prompt word templates, construction data review prompt word templates, risk verification prompt word templates, etc. Business processing rules can include specification clause matching rules, template filling rules, field permission rules, data query rules, risk verification rules, review report structure rules, etc. The large model can be a large language model deployed within the enterprise and can provide unified services to business systems through the model service interface.

[0026] After generating the initial processing results, consistency verification and source tracing are performed on the initial processing results based on the processing strategy, target object relationships, and target source identifiers to obtain the target processing results. Consistency verification can include factual consistency verification, field consistency verification, standard basis consistency verification, business relationship consistency verification, permission consistency verification, and output format consistency verification. Source tracing can bind key conclusions, cited content, problem points, rectification suggestions, and analysis indicators in the target processing results to the corresponding target source identifiers, allowing users to view the documents, records, standard clauses, data tables, or business process nodes upon which the target processing results are based. Finally, the target processing results are output. The output format can be at least one of the following: Q&A results, data analysis results, supervision document content, document review results, and risk verification results. It can also be further exported as a document, table, chart, or structured report.

[0027] Through the above implementation methods, supervision business data scattered across different systems, documents, and business processes can be parsed into a supervision business knowledge index with correlation and source identification. Based on the target supervision business request, the business processing type and processing strategy can be automatically determined. Then, by combining the large model, prompt word templates, and business processing rules, verifiable and traceable target processing results can be generated. This improves the efficiency of supervision business data utilization, reduces the reliance on human experience in the supervision data processing and review process, and enhances the comprehensive identification capability of on-site risk information.

[0028] In one implementation, when performing format parsing and semantic extraction on multi-source supervision business data, the data format and data source of the multi-source supervision business data are first identified to obtain data format identifiers and data source identifiers. The data format identifier can be used to characterize whether the data to be processed belongs to a text file, spreadsheet file, presentation file, web page file, image file, process record, database record, or interface data. The data source identifier can be used to characterize whether the data originates from a project data management system, a supervision mobile application, a document library, a business database, locally uploaded files, FTP uploaded files, on-site data acquisition equipment, or a third-party interface.

[0029] Based on the data format identifier, the corresponding parsing strategy is invoked to parse the content of multi-source supervision business data. For document data, it supports parsing formats such as pdf, doc, docx, ppt, pptx, mhtml, xlsx, xls, md, txt, csv, and html; for image data, it supports parsing formats such as jpg, png, and jpeg. For files with complex layouts including mixed text, images, tables, nested tables, formulas, images, titles, paragraphs, headers, and footers, a multimodal document parsing model can be used for coarse-grained layout element location and reading order recognition. This is then combined with text recognition, table structure recovery, formula recognition, and image description generation to convert the file content into standardized business content. Standardized business content can include at least one of the following: text content, table content, image content, title content, paragraph content, header / footer content, and process record content.

[0030] After obtaining standardized business content, it is reordered according to its layout, content level, and contextual semantics to obtain the business content to be extracted, conforming to the business reading order. For example, chapter titles, body paragraphs, table data, and diagrams in construction plans can be restored to their reading order according to chapter level and page layout; project overviews, progress, quality and safety status, and problem rectification status in monthly supervision reports can be reordered according to template chapters and business semantics; and process approval records can be formed into business content to be extracted according to the process node order, processing time, and processing role.

[0031] After obtaining the business content to be extracted, entity recognition and attribute extraction are performed on the business content to obtain the business objects and their corresponding object attributes. Entity recognition can be used to identify project names, contract names, construction units, supervision units, personnel names, certificate numbers, construction content, risk items, quality issues, safety hazards, and standard clauses, etc. Attribute extraction can be used to extract attributes such as time, location, status, number, type, approval conclusion, validity period, rectification period, and risk level corresponding to the business objects. For table content, object attributes can be determined by combining table headers, row and column relationships, and cell content; for image content, business objects related to the image can be determined by combining image recognition results and image context description.

[0032] After obtaining the business objects and their attributes, the object relationships between business objects are determined based on the contextual position of the business object within the business content to be extracted and the matching relationship between the object attributes. The data source identifier is then bound to the business object to obtain the source identifier. For example, the relationship between different business objects can be determined based on the same chapter, table, process record, or project number; the relationship between personnel objects and qualification objects can be determined based on personnel name, certificate number, ID information, entry records, etc.; and the relationship between construction data objects and standard reference objects can be determined based on construction content, specification clauses, and review opinions. The source identifier can be refined to file name, file page number, paragraph number, table row / column position, image number, process node number, database record number, or equipment acquisition record number.

[0033] Through the above implementation methods, supervision business data of different formats, sources, and complex layouts can be uniformly parsed and semantically extracted. The parsed business objects, object relationships, and source identifiers are formed into basic data that can be used for subsequent index construction and business processing, thereby reducing the workload of manually organizing documents and manually identifying business relationships, and improving the structuring degree and traceability of multi-source supervision business data.

[0034] In one implementation, when constructing a supervision business knowledge index based on business objects, object relationships, and source identifiers, the business objects are first classified according to their object attributes, resulting in at least one of the following categories: project objects, document objects, personnel objects, quality and safety objects, process objects, and standard basis objects. Project objects may include engineering projects, sections, unit projects, and sub-items of engineering projects; document objects may include construction plans, supervision plans, supervision details, supervision logs, monthly supervision reports, acceptance reports, and project schedules; personnel objects may include operators, supervisors, construction unit personnel, and special operation personnel; quality and safety objects may include quality problems, safety hazards, risk events, rectification measures, and acceptance conclusions; process objects may include reporting and approval processes, approval nodes, processing opinions, and process status; and standard basis objects may include national industry standards, enterprise technical standards, management systems, and review clauses.

[0035] After obtaining the classification objects, the classification objects and their corresponding standardized business content are vectorized to obtain vector indexes. Vectorization can employ text embedding models or domain-adapted embedding models to convert business object names, object attributes, contextual content, regulatory clauses, question-answer pairs, etc., into vector representations, which are then stored in a vector database or vector retrieval component. For specialized domain questions, the embedding representations can be optimized based on power engineering supervision terminology, regulatory clauses, and internal enterprise business corpora to improve the accuracy of semantic retrieval.

[0036] Keyword indexing is generated by extracting keywords and generating tags from categorized objects and their corresponding attributes. Keywords can include project names, document types, personnel types, certificate types, risk types, specification names, clause numbers, process nodes, and business statuses. Tags can be generated manually, automatically, or a combination of both. For documents uploaded to the knowledge base, document summaries and tags can be automatically generated after parsing, or users can select existing tags or manually add tags. The keyword indexing supports fast retrieval, precise matching, tag filtering, and statistics on high-frequency business keywords.

[0037] Based on object relationships, a structured association index is generated to represent the business relationship paths between different categories of objects. This structured association index can be implemented using graph structures, relational tables, inverted indexes, or knowledge graphs to describe the relationship paths between objects such as projects, documents, personnel, qualifications, risks, standards, and processes. For example, a worker can be associated with entry approval records, safety regulation examination records, special operation qualification certificates, violation records, and on-site identification records; a construction plan can be associated with construction content, chapter elements, review basis, review issues, and rectification suggestions. The structured association index supports graph retrieval, path retrieval, association analysis, and multi-hop reasoning.

[0038] After constructing various indexes, mapping relationships are established between source identifiers and vector indexes, keyword indexes, and structured association indexes to obtain the supervision business knowledge index. The supervision business knowledge index can be uniformly managed through a knowledge base management tool, supporting operations such as adding knowledge bases, viewing knowledge base details, uploading documents, configuring domains, configuring vector processing methods, managing knowledge base online / offline status, viewing document lists, managing question-and-answer pairs, generating document summaries, splitting question-and-answer pairs, configuring text embedding strategies, and configuring vector storage methods. Furthermore, the accuracy of the supervision business knowledge index can be continuously optimized through dialogue feedback, data revision, and data feedback mechanisms.

[0039] Through the above implementation methods, scattered engineering supervision data, business records, personnel information, quality and safety data, and standard references can be uniformly organized into a supervision business knowledge index that includes vector indexes, keyword indexes, and structured association indexes. This enables supervision business knowledge to be semantically retrieved, precisely matched with keywords, and traced along relationship paths, thereby improving the retrieval accuracy and data utilization efficiency of professional Q&A, data analysis, document review, and risk verification.

[0040] In one implementation, when identifying the intent of a target supervision business request, semantic parsing can be performed first to obtain the request action, request object, query conditions, and result requirements. Request actions may include querying, statistics, analysis, generation, polishing, summarizing, reviewing, verifying, and modification suggestions; request objects may include project materials, business data, supervision documents, construction data, personnel, quality and safety issues, and regulatory basis; query conditions may include project scope, time range, document type, personnel type, process status, and risk level; result requirements may include response format, chart type, export format, review report structure, summary length, level of professionalism, and whether supporting documentation is required.

[0041] Based on the request action and the request object, determine the corresponding business processing type for the target supervision business request. When the target supervision business request is mainly used to obtain professional knowledge, query project information, or view historical dialogue content, it can be identified as an information query type; when the target supervision business request is used to query business data, statistical indicators, generate charts, or export analysis results, it can be identified as a data analysis type; when the target supervision business request is used to generate documents such as supervision reports, notices, meeting minutes, monthly supervision reports, supervision logs, acceptance reports, and project schedules, it can be identified as a document generation type; when the target supervision business request is used to review construction plans, construction data, or other approval documents, it can be identified as a document review type; when the target supervision business request is used to verify the identity, qualifications, approval, examination, or violation risks of on-site personnel, it can be identified as a risk verification type.

[0042] Based on the query conditions and result requirements, determine the search scope, search method, output format, and validation rules corresponding to the business processing type. The search scope can be limited to a specified project, knowledge base, document type, personnel range, or time range. Search methods can include vector search, keyword search, structured association search, and combinations of multiple search methods. Output formats can include natural language text, structured tables, charts, professional documents, review reports, risk warnings, etc. Validation rules can include source consistency verification, field permission verification, standard clause matching verification, content integrity verification, format specification verification, and identity matching confidence level verification, etc.

[0043] Based on business processing type, search scope, search method, output format, and validation rules, a processing strategy is generated. This strategy can be further bound to corresponding prompt word templates, system role definitions, model parameters, service interfaces, permission levels, and audit requirements. Prompt word templates can be uniformly maintained by the prompt word management module, supporting filtering, editing, application, and deletion by business type, and the scope of responses can be narrowed through system role definitions. The model service provides unified invocation capabilities through service management, model version configuration, service scheduling configuration, and service API authentication management. When target supervision business requests involve modifications to business system data or process operations, permission control policies can be generated based on user roles, department permissions, and operation types, requiring operation logs to be recorded.

[0044] Through the above implementation methods, the target supervision business request in natural language form or after speech conversion can be parsed into a task with a clear business processing type and processing strategy. This ensures that subsequent retrieval, model invocation, result generation, verification, and output are all constrained by the business scenario, retrieval scope, output format, and verification rules, thereby reducing the deviation in request understanding and improving the accuracy, controllability, and security of intelligent processing of power engineering supervision business.

[0045] In one implementation, when the business processing type is information query, candidate business objects and their relationships are retrieved from at least one of the following indexes: vector index, keyword index, and structured association index, based on the retrieval scope and retrieval method in the processing strategy. Information query requests may include user-submitted questions about power engineering, project data, regulatory standards, or follow-up questions from past conversations. Users can input questions via natural language or voice; the system can first convert the voice into text and then initiate a retrieval based on the target supervision business request.

[0046] When retrieving candidate business objects and their relationships, the system can retrieve semantically similar document fragments, specification clauses, question-and-answer pairs, or historical dialogues from the vector index based on the semantic vector of the target supervision business request; it can perform precise matching from the keyword index based on keywords, tags, clause numbers, project names, etc., in the target supervision business request; and it can perform association path retrieval from the structured association index based on the relationship between the requested object and objects such as projects, documents, specifications, personnel, and processes. For question-and-answer pair corpora already existing in the knowledge base, a high-priority matching strategy can be set to prioritize standard question-and-answer pairs in candidate result generation.

[0047] After obtaining the candidate business objects and their relationships, the candidate business objects and their relationships are ranked based on the semantic relevance, keyword matching degree, and association path matching degree between the candidate business objects and the target supervision business requests, thus obtaining the target business objects and their relationships. Semantic relevance can be determined based on vector similarity; keyword matching degree can be determined based on the number of keyword hits, tag matching, and clause number matching; and association path matching degree can be determined based on the association path length, association type, and business credibility between business objects. This comprehensive ranking reduces the amount of irrelevant data entering the subsequent large-scale model processing.

[0048] Based on the source identifier corresponding to the target business object and its associated relationships, a target source identifier is determined. Then, a question-and-answer context is generated based on the target business object, its associated relationships, and the target source identifier to answer the target supervision business request. The question-and-answer context may include factual content corresponding to the target business object, regulatory basis, source location, relevant historical dialogues, content from adjacent chapters, and necessary answer constraints. If the target supervision business request is a follow-up question in a multi-turn dialogue, the question-and-answer context can also be generated by combining historical dialogue summaries, user feedback records, and the selected knowledge base scope.

[0049] The system inputs the target supervision business request and the question-and-answer context into the large model to obtain the question-and-answer results, which are then used as the initial processing results. The question-and-answer results can include direct answers, supporting content, source annotations, and related knowledge recommendations. The system also supports dialogue status management, such as generating, generating, and generating completed states; it supports previewing, continuing question-and-answer sessions, deleting, editing, pinning, and unpinning historical dialogues; and it supports user feedback on the question-and-answer results, with feedback entering the data feedback flow for continuous calibration of the knowledge base and model performance. The system can also display data statistics on "My Questions and Answers," a high-frequency question list, recently used or favorited knowledge base entries, a business hot keyword list, and general dialogue and knowledge base question-and-answer data through a dashboard.

[0050] Through the above implementation methods, when users raise professional questions about power engineering supervision or inquire about project information, the system can retrieve relevant and traceable business objects, relationships, and source identifiers from the supervision business knowledge index. Combined with a large model, it can generate professional and reliable question-and-answer results, thereby improving the efficiency of professional information retrieval, reducing the burden on supervisors to manually retrieve information and make judgments, and enhancing the interpretability and traceability of the question-and-answer results.

[0051] In one implementation, when the business processing type is data analysis, the business objects to be analyzed, analysis indicators, analysis dimensions, filtering conditions, and result display requirements are determined based on the target supervision business request. The business objects to be analyzed can be engineering projects, quality issues, safety hazards, personnel qualifications, rectification records, process approval items, etc.; analysis indicators can be quantity, proportion, completion rate, rectification rate, overdue rate, risk level distribution, etc.; analysis dimensions can be project, unit, time, professional category, risk type, process status, etc.; filtering conditions can be time range, project scope, document type, personnel type, problem status, etc.; result display requirements can be text summary, table, bar chart, line chart, pie chart, exported file, etc.

[0052] Based on the business objects to be analyzed, the analysis indicators, the analysis dimensions, and the filtering conditions, the target data tables, target fields, and field relationships are determined from the structured relational index. The structured relational index records the correspondence between business objects and underlying data tables, fields, and interface parameters. For example, quality and safety objects can correspond to quality problem tables, safety hazard tables, rectification record tables, etc.; personnel objects can correspond to personnel basic information tables, entry approval forms, examination record tables, qualification certificate tables, etc.; process objects can correspond to approval node tables, process status tables, and operation log tables, etc. Field relationships can include primary key associations, project number associations, personnel number associations, process instance number associations, etc.

[0053] Based on the target data table, target fields, and field relationships, a business data query statement is generated. During the generation of the business data query statement, a large model combined with prompt word templates can be used to convert natural language into query statements. Alternatively, vector retrieval, structured association retrieval, and keyword retrieval can be combined to enhance the model's understanding of table information, field information, association information, and conditional information. The generated business data query statement can be displayed to users or administrators for manual verification. During the generation process, the query request and corresponding query statement can also be vectorized and stored as a basis for subsequent reuse of similar queries and optimization of query statements.

[0054] After generating the business data query statement, it undergoes syntax validation, field permission validation, and query scope validation to obtain a valid query statement. Syntax validation determines if the query statement conforms to database syntax requirements; field permission validation determines if the current user has permission to access the target fields and target data tables; and query scope validation determines if the query conditions exceed the allowed project, department, or time range. If the business data query statement fails validation, it is corrected based on the reason for failure, resulting in a revised query statement. Correction methods may include regenerating the query statement, adjusting field names, adding query conditions, deleting fields with insufficient permissions, and prompting for manual correction.

[0055] The system executes validated or modified business data query statements to obtain business query results. Based on the business query results and display requirements, it generates at least one of the following: data analysis text, chart configuration information, and data export information, which serves as the initial processing result. The data analysis text can summarize the query results; the chart configuration information can include chart type, horizontal and vertical axes, data series, legend, title, and interaction configurations; and the data export information can be used to generate CSV, Excel, PDF, or other files. The chart type can be automatically determined by the system based on data field types, analysis dimensions, and user intent, or it can be re-rendered and optimized based on user adjustment requests during multi-turn dialogues. The chart data can be converted into the data format and configuration required by the front-end chart components to support dynamic display and interaction.

[0056] Through the above implementation methods, the data query and analysis requirements proposed by users in natural language can be converted into verifiable and executable business data query statements. Under the condition of controlled permissions, business query results, analysis text, chart configuration and export information can be generated, thereby reducing the difficulty of manually writing query conditions and business reports, improving the efficiency of supervision business data analysis, and reducing the data security risks caused by erroneous queries or unauthorized queries.

[0057] In one implementation, when the business processing type is document generation, the document type and purpose of the document to be generated are determined based on the target supervision business request. Document types may include supervision reports, work notices, meeting minutes, monthly supervision reports, supervision logs, acceptance reports, project schedules, supervision plans, supervision details, and supervision work documents. Document purposes may include daily business communication, project reporting, document archiving, acceptance support, problem rectification notices, and meeting minutes compilation. The target supervision business request may also include requirements for length, tone, professionalism, and output format.

[0058] Based on the document type and purpose, the corresponding power engineering supervision document template is obtained, resulting in the target document template. The target document template can be a predefined standard template, including document title, chapter structure, field items, table items, issuance information, body text format, terminology style, etc. Different types of target document templates can correspond to professional documents such as monthly supervision reports, supervision logs, acceptance reports, project schedules, supervision plans, and supervision details. The target document template can also be bound to power industry standards, internal company regulations, and common writing style requirements to ensure that the generated content conforms to the format specifications and industry language style of power engineering supervision business.

[0059] Based on the fields and sections in the target document template, relevant project factual information, process record information, quality and safety information, and regulatory basis content are extracted from the business context to obtain template population information. Project factual information may include project name, construction unit, construction company, supervision unit, project progress, key milestones, etc.; process record information may include approval records, submission status, processing opinions, rectification closure status, etc.; quality and safety information may include quality problems, safety hazards, risk items, rectification measures, and supervision conclusions, etc.; regulatory basis content may include standard clauses, system requirements, and management regulations related to the document content. For fields in the target document template with missing content, adjacent business objects can be retrieved through related indexes, or supplementary suggestions can be generated.

[0060] The template fill information is mapped to corresponding source identifiers and field / section items in the target document template, thus creating a field source mapping relationship. This mapping relationship records the sources of project materials, business records, specification clauses, table cells, process nodes, etc., referenced by each field or section item. Based on this mapping relationship, the template fill information is then filled into the target document template to obtain the draft document content. During the filling process, date formats, organization names, project numbers, data indicators, table styles, and section numbers can be uniformly processed.

[0061] Based on the formatting requirements, content completeness requirements, and field source mapping relationships corresponding to the target document template, the draft document content is completed and standardized to obtain the supervision document content, which is then used as the initial processing result. Completion and standardization processes may include adding transitional phrases, adjusting technical terms, polishing expressions, deleting redundant content, generating summaries, extracting key conclusions, standardizing formats, and verifying field completeness. For general content creation scenarios, reports, notices, meeting minutes, and other texts can be generated; for word processing scenarios, polishing, summarizing, and professional expression enhancement can be performed; for professional document generation scenarios, a standardized initial draft of a professional document can be generated strictly according to the target document template.

[0062] Through the above implementation method, the document type and purpose can be automatically determined according to the target supervision business request, and the supervision document content can be generated based on the power engineering supervision document template, business context and field source mapping relationship. The generated document not only conforms to the supervision business format and industry language style, but also allows for the traceability of key content sources, thereby reducing the workload of manual writing and manual modification, and improving the efficiency of professional document generation and content reliability.

[0063] In one implementation, when the business processing type is document review, the construction documents to be reviewed are determined from the target supervision business request or multi-source supervision business data. The construction documents to be reviewed can include construction plans, specialized construction plans, submission documents, technical measures documents, safety measures documents, schedule documents, acceptance documents, etc. The construction documents to be reviewed can be uploaded locally, selected from a document library, automatically passed in from process nodes, or obtained through business system interfaces.

[0064] After identifying the construction documents to be reviewed, chapter identification and content element extraction are performed to obtain structured content. Chapter identification can identify titles, chapter levels, table of contents, body paragraphs, tables, formulas, images, attachments, etc. Content element extraction can extract elements such as project overview, construction content, personnel allocation, machinery and equipment, construction technology, safety measures, quality control measures, schedule, emergency plan, calculations, and acceptance standards. For construction documents with mixed text, graphics, and tables, or containing nested tables, document parsing and reading order reorganization can be performed first, followed by element extraction.

[0065] Based on the structured content of the construction documents, compliance review items, completeness review items, matching degree review items, and risk logic review items are determined. Compliance review items are used to determine whether the content of the construction documents complies with national industry standards, enterprise technical standards, and management systems; completeness review items are used to determine whether the construction documents lack key modules such as safety measures, quality control, schedule plans, personnel allocation, and emergency measures; matching degree review items are used to determine whether the construction content is consistent with the scheme description, submitted content, and project documents; risk logic review items are used to identify potential safety risks, quality risks, logical contradictions, and calculation errors.

[0066] Based on compliance review items, completeness review items, matching degree review items, and risk logic review items, the corresponding review basis is retrieved from the supervision business knowledge index to obtain the target review basis. The target review basis may include national industry standards, enterprise technical standards, management systems, historical review opinions, similar project review records, and professional knowledge base content. Search methods may include keyword search, vector search, and structured association search. For example, corresponding standard clauses can be retrieved based on construction process keywords, similar document review requirements can be retrieved based on chapter semantics, and enterprise standard clauses can be linked based on project type and construction content.

[0067] Based on the target review criteria, the structured content of the construction documents was subjected to several checks, including matching with regulatory clauses, detecting missing key modules, comparing the consistency of construction content, and verifying risk logic, to identify problems. Matching with regulatory clauses determines whether the technical measures, acceptance standards, and operational requirements in the construction documents are consistent with the target review criteria; detecting missing key modules determines whether the construction documents are missing necessary chapters or key content; comparing the consistency of construction content compares the construction content in the documents with project documents, approval information, and site records; and verifying risk logic identifies problems such as inconsistencies in descriptions, insufficient risk measures, and abnormal calculation data.

[0068] Based on the problem identification results and the target review criteria, a structured review result containing problem points, supporting content, and rectification suggestions is generated, and this structured review result is used as the initial processing result. The structured review result can be organized in the form of problem points, supporting content, and rectification suggestions, and can be annotated with corresponding regulatory clauses, source documents, chapter positions, and severity levels. For review items without problems, a review pass conclusion can be recorded; for review items with problems, specific rectification suggestions and required supplementary materials can be provided. The structured review result can be exported as a review report or written into business process nodes for subsequent rectification closure.

[0069] Through the above implementation methods, construction data can be parsed into structured content, and based on the standard basis and business data in the supervision business knowledge index, a multi-dimensional review can be conducted on compliance, completeness, matching degree, and risk logic. This reduces the workload of manually checking standard clauses and data content item by item, and improves the consistency, accuracy, and problem location efficiency of construction data review.

[0070] In one implementation, when the business processing type is risk verification, on-site identification data collected by on-site data acquisition devices is obtained. On-site data acquisition devices may include recorders, cameras, mobile terminals, access control devices, or other IoT devices. On-site identification data may include on-site images, video frames, facial images, document images, device acquisition time, acquisition location, and acquisition device serial number. On-site identification data can be actively collected by on-site supervisors or automatically collected by the business system during personnel entry, on-site inspections, or work patrols.

[0071] The on-site identification data is used to identify objects to be verified, including personnel. When identifying personnel, facial recognition services can be invoked to detect, evaluate quality, extract features, and compare faces in on-site images or video frames. Capabilities such as document OCR recognition, liveness detection, enhanced device security, enhanced liveness security, and enhanced intelligent hierarchical authentication can be combined to improve the reliability of identity verification. For scenarios involving mask-wearing, facial recognition capabilities supporting occlusion can be used. Abnormal identification results can be reviewed through an intelligent audit process.

[0072] The process involves matching the individuals to be verified against their baseline information to obtain the identity matching result and the confidence level. The baseline information can be sourced from a personnel database, entry approval records, company personnel database, document information database, or on-site personnel registration database. The confidence level is determined comprehensively based on factors such as facial recognition score, document OCR consistency, liveness detection results, equipment security testing results, and personnel database match rates. If the confidence level meets preset identity verification conditions, the system retrieves corresponding personnel information, entry approval information, qualification certificate information, examination record information, and violation record information from the supervision business knowledge index based on the identity matching result, thus obtaining the associated data for the on-site individuals.

[0073] Based on the associated data of the on-site personnel, determine the identity status, application status, qualification status, examination status, and historical risk status of the personnel to be verified. Identity status indicates whether the personnel's identity matches the description, whether there is impersonation, or whether the documents are abnormal; application status indicates whether the personnel have completed the on-site application process; qualification status indicates whether the special operation qualification is valid, expired, or matches the type of operation; examination status indicates whether the personnel have passed the safety regulations examination; and historical risk status indicates whether the personnel have any violations, historical safety risks, or other abnormal records.

[0074] The system matches identity status, application status, qualification status, examination status, and historical risk status against preset risk verification rules to obtain risk hit results. Preset risk verification rules may include rules for personnel entering the site without approval, unlicensed work, expired qualifications, forged documents, failed examinations, and high-risk historical violations. For hit risk rules, corresponding risk types, risk levels, hitting basis, handling suggestions, and source identifiers are generated. Based on the risk hit results, risk verification results are generated and used as the initial processing results. Risk verification results can be displayed on the front-end page or pushed to on-site supervisors, project managers, or the risk rectification process.

[0075] Regarding data security and compliance, personnel identity data, facial data, and document data can be processed in accordance with personal information protection requirements. The system supports hierarchical management of the facial database, access control, log auditing, data anonymization, verification record retention, and compliance assessment. For large-scale personnel database scenarios, response speed can be improved through facial indexing, feature retrieval, and hierarchical authentication, and model self-learning and intelligent review can be supported to continuously improve recognition accuracy.

[0076] Through the above implementation methods, the on-site identification data acquired by the on-site data collection equipment can be correlated and verified with personnel information, entry approval information, qualification certificate information, examination record information, and violation record information. This automatically identifies risks such as personnel entering the site without approval, working without a certificate, expired qualifications, abnormal certificates, and historical violations, thereby improving the timeliness of on-site risk identification and the risk control capabilities of engineering supervision.

[0077] In one implementation, a large-scale model-based intelligent processing system for power engineering supervision is provided. This system is deployed on a supervision business management platform and / or a mobile supervision application. The system can be implemented using a web application architecture and a front-end / back-end separation approach. The PC client can run on a Windows environment, and the mobile client can run on an iOS or Android environment. The system's software implementation can use languages ​​such as Java, JavaScript, CSS, and HTML. The back-end can use frameworks such as Tomcat, Spring Boot, and Camunda, while the front-end can use frameworks such as Vue, ElementUI, ECharts, and UniApp. The system can adhere to unified code standards, standardized interface standards, and unified user interface standards, enabling the large-scale model capabilities to be embedded into multiple stages of the supervision business process.

[0078] The system may include a data acquisition module, a parsing and extraction module, an index building module, an intent recognition module, a context generation module, a large model processing module, a verification and tracing module, and a result output module. The data acquisition module acquires the target supervision business request and related multi-source supervision business data. The parsing and extraction module performs format parsing and semantic extraction on the multi-source supervision business data to obtain business objects, object relationships, and source identifiers. The index building module constructs a supervision business knowledge index based on business objects, object relationships, and source identifiers. The intent recognition module determines the business processing type and processing strategy corresponding to the target supervision business request. The context generation module retrieves the target business object, target object relationships, and target source identifier, and generates a business context. The large model processing module calls the prompt word template and / or business processing rules, inputs the target supervision business request and business context into the large model, and generates initial processing results. The verification and tracing module performs consistency verification and source tracing annotation on the initial processing results to obtain the target processing result. The result output module outputs the target processing result.

[0079] The system can also include knowledge base management, knowledge library management, intelligent question answering, dialogue management, prompt word management, model service management, and system management functions. The knowledge base management function supports knowledge base retrieval, adding new knowledge bases, configuring knowledge base names, descriptions, domains, and vector processing methods, managing knowledge base online / offline status, and viewing knowledge base lists and details. The knowledge library management function supports local uploads, FTP uploads, library tag management, automatic library summary generation, automatic question-and-answer pair generation, question-and-answer pair splitting, and file processing configuration selection. The intelligent question answering function supports both knowledge base question answering and model question answering modes, and supports domain models, tagged knowledge bases, question answering strategies, source preview, voice input, prompt word template enhancement, and question-and-answer feedback. The dialogue management function supports historical dialogue retrieval, general dialogue list management, knowledge base question answering dialogue list management, feedback dialogue management, data revision, saving, and publishing. The prompt word management function supports the classification, editing, application, and deletion of prompt word templates. The model service management function supports unified external service provision, service list management, model and version selection, model parameter configuration, service performance testing, service API management, and authentication management. The system management function supports theme switching, personal information management, password modification, department management, and user management.

[0080] The system can also include intelligent query and analysis functions. This function supports both text and voice input, enhancing natural language query understanding based on vector retrieval, structured association retrieval, and keyword retrieval. It converts natural language queries into table information, field information, association information, and conditional information, generating business data query statements. The system can display these business data query statements, supporting syntax validation, permission validation, manual correction, regeneration, and standardized management. It also supports query result display and export in formats such as CSV, Excel, and PDF. For chart presentation, the system can automatically generate bar charts, line charts, pie charts, etc., based on the query results. It supports adjusting chart styles through multi-turn dialogues, converting chart data into the data format and configuration required by the front-end chart components for dynamic display and interaction.

[0081] The system can also include document generation and document review functions. Document generation supports general content creation, text polishing, summary generation, and professional document generation, capable of generating supervision reports, work notices, meeting minutes, monthly supervision reports, supervision logs, acceptance reports, project schedules, supervision plans, supervision details, and supervision work documents. The document review function, relying on industry standards, enterprise standards, and a supervision business knowledge index, conducts compliance, completeness, matching, and risk logic reviews of construction documents, and outputs structured review results organized according to problem points, supporting content, and rectification suggestions.

[0082] The system can also include controlled system interaction functions and risk verification functions. The controlled system interaction function can receive AI commands for specific business functions and, under conditions of hierarchical access control, user authentication, and operation log recording, execute data modifications or generate modification suggestions. The operation log can record the operation initiator, operation time, content before modification, content after modification, and operation type, and supports filtering and querying by time, operator, and data type, meeting audit traceability requirements. The risk verification function can connect to on-site data collection devices and, combined with capabilities such as facial recognition, document OCR recognition, liveness detection, facial comparison, enhanced device security, enhanced liveness security, enhanced intelligent hierarchical authentication, multimodal models, abnormal template recognition, model self-learning, and intelligent review, dynamically verify on-site personnel, identifying risks such as unapproved entry, unlicensed work, expired qualifications, abnormal documents, and historical violations.

[0083] The system can be deployed on domestically produced hardware platforms to form a large-scale model basic inference service capability, supporting dynamic expansion of computing resources. The hardware platform can include a training and inference integrated machine and support mainstream domestic operating systems. The training and inference integrated machine can be configured with multiple high-core CPUs, large-capacity memory, multiple computing modules, large-capacity storage, gigabit Ethernet ports, and high-speed optical ports to meet the needs of document parsing, model training, and model inference. The system also possesses full lifecycle capabilities for large models, including large-scale model management, large-scale model data management, large-scale model training, large-scale model inference, large-scale model training and inference computing resource management, monitoring, and scheduling, thereby supporting enterprise-level private deployment and business system integration.

[0084] Through the above implementation methods, the system can integrate large model capabilities, supervision business data, engineering knowledge base, document parsing capabilities, intelligent data query capabilities, professional document generation capabilities, document review capabilities, and risk verification capabilities on the basis of a unified enterprise-level software interface, standardized interfaces, access control, and computing resource management. This enables information query, data analysis, document generation, document review, and on-site risk verification in supervision business to be processed intelligently, standardized, and traceably within the same business system, thereby improving the efficiency of engineering supervision business processing and risk control capabilities.

[0085] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for intelligent processing of power engineering supervision business based on a large model, characterized in that, The method includes: Obtain the target supervision business request and multi-source supervision business data related to the target supervision business request. The multi-source supervision business data includes at least one of project data, on-site record data, personnel information data, quality and safety data, and management process data. The multi-source supervision business data is parsed and semantically extracted to obtain business objects, object relationships between business objects, and source identifiers used to characterize the source of the business objects; Based on the business object, the object association, and the source identifier, a supervision business knowledge index is constructed, which includes a vector index, a keyword index, and a structured association index. The intent of the target supervision business request is identified to determine the business processing type and processing strategy corresponding to the target supervision business request. The business processing type includes at least one of information query, data analysis, document generation, document review and risk verification. Based on the business processing type and the processing strategy, retrieve the target business object, target object association, and target source identifier that match the target supervision business request from the supervision business knowledge index, and generate a business context based on the target business object, the target object association, and the target source identifier; Based on the business processing type, the corresponding prompt word template and / or business processing rule are invoked, the target supervision business request and the business context are input into the large model, and the initial processing result is generated. Based on the processing strategy, the target object association, and the target source identifier, the initial processing result is subjected to consistency verification and source tracing annotation to obtain the target processing result; Output the target processing result; The step of constructing a supervision business knowledge index based on the business object, the object association, and the source identifier includes: Based on the object attributes of the business object, the business object is classified into business categories to obtain at least one of the following categories: project object, document object, personnel object, quality and safety object, process object, and standard basis object. The classification objects and the standardized business content corresponding to the classification objects are vectorized to obtain vector indexes; Keyword index is obtained by extracting keywords and generating tags from the classification objects and their corresponding object attributes. Based on the object association relationships, a structured association index is generated to represent the business association paths between different categories of objects; The source identifier is mapped to the vector index, the keyword index, and the structured association index respectively to obtain the supervision business knowledge index.

2. The intelligent processing method for power engineering supervision business based on a large model according to claim 1, characterized in that, The step of parsing and semantically extracting the multi-source supervision business data to obtain business objects, object relationships between business objects, and source identifiers used to characterize the source of the business objects includes: Identify the data format and data source of the multi-source supervision business data to obtain data format identifier and data source identifier; Based on the data format identifier, the corresponding parsing strategy is invoked to parse the multi-source supervision business data to obtain standardized business content that includes at least one of the following: text content, table content, image content, title content, paragraph content, header and footer content, and process record content. Based on the layout, content level, and contextual semantics of the standardized business content, the standardized business content is reorganized in order to obtain business content to be extracted that conforms to the business reading order. Entity recognition and attribute extraction are performed on the business content to be extracted to obtain the business object and the object attributes corresponding to the business object. Based on the matching relationship between the context position of the business object in the business content to be extracted and the object attribute, the object association relationship between the business objects is determined, and the data source identifier is bound to the business object to obtain the source identifier.

3. The intelligent processing method for power engineering supervision business based on a large model according to claim 1, characterized in that, The step of performing intent recognition on the target supervision service request and determining the corresponding service processing type and processing strategy includes: The target supervision business request is semantically parsed to obtain the request action, request object, query conditions and result requirements; Based on the request action and the request object, determine the business processing type corresponding to the target supervision business request; Based on the query conditions and the result requirements, determine the retrieval scope, retrieval method, output format, and verification rules corresponding to the business processing type; The processing strategy is generated based on the business processing type, the search scope, the search method, the output format, and the verification rules.

4. The intelligent processing method for power engineering supervision business based on a large model according to claim 1, characterized in that, When the business processing type is information query, the step of retrieving the target business object, target object association, and target source identifier that match the target supervision business request from the supervision business knowledge index according to the business processing type and the processing strategy, and generating a business context based on the target business object, the target object association, and the target source identifier, includes: Based on the retrieval scope and retrieval method in the processing strategy, candidate business objects and candidate object associations are retrieved from at least one of the vector index, the keyword index, and the structured association index. Based on the semantic relevance, keyword matching degree, and association path matching degree between the candidate business objects and the target supervision business request, the candidate business objects and their association relationships are sorted to obtain the target business objects and their association relationships. The target source identifier is determined based on the source identifier corresponding to the target business object and the association relationship of the target object; Based on the target business object, the target object's association relationship, and the target source identifier, a question-and-answer context is generated to answer the target supervision business request; The target supervision business request and the question-and-answer context are input into the large model to obtain the question-and-answer results, and the question-and-answer results are used as the initial processing results.

5. The intelligent processing method for power engineering supervision business based on a large model according to claim 1, characterized in that, When the business processing type is data analysis, the step of calling the corresponding prompt word template and / or business processing rule according to the business processing type, inputting the target supervision business request and the business context into the large model, and generating initial processing results includes: Based on the target supervision business request, determine the business object to be analyzed, the analysis indicators, the analysis dimensions, the screening conditions, and the result display requirements; Based on the business object to be analyzed, the analysis indicators, the analysis dimensions, and the filtering conditions, the target data table, target fields, and field relationships are determined from the structured association index. Based on the target data table, the target fields, and the relationships between the fields, a business data query statement is generated; The business data query statement is subjected to syntax validation, field permission validation, and query range validation to obtain a valid business data query statement. If the business data query statement fails the validation, the business data query statement is corrected according to the reason for the validation failure, resulting in a corrected business data query statement. Execute the verified business data query statement or the corrected business data query statement to obtain the business query result; Based on the business query results and the result display requirements, at least one of the following is generated: data analysis text, chart configuration information, and data export information. At least one of the following is used as the initial processing result: data analysis text, chart configuration information, and data export information.

6. The intelligent processing method for power engineering supervision business based on a large model according to claim 1, characterized in that, When the business processing type is document generation, the step of calling the corresponding prompt word template and / or business processing rule according to the business processing type, inputting the target supervision business request and the business context into the large model, and generating the initial processing result includes: Based on the target supervision service request, determine the document type and purpose of the document to be generated; Based on the document type and the document purpose, obtain the corresponding power engineering supervision document template to obtain the target document template; Based on the field items and chapter items in the target document template, the corresponding project fact information, process record information, quality and safety information, and standard basis content are extracted from the business context to obtain template filling information; Establish a correspondence between the source identifier corresponding to the template fill information and the field items and chapter items in the target document template to obtain the field source mapping relationship; Based on the field source mapping relationship, the template filling information is filled into the target document template to obtain the document draft content; Based on the format requirements, content integrity requirements, and field source mapping relationships corresponding to the target document template, the content of the document draft is completed and standardized to obtain the content of the supervision document, and the content of the supervision document is used as the initial processing result.

7. The intelligent processing method for power engineering supervision business based on a large model according to claim 1, characterized in that, When the business processing type is document review, the step of calling the corresponding prompt word template and / or business processing rule according to the business processing type, inputting the target supervision business request and the business context into the large model, and generating the initial processing result includes: The construction documents to be reviewed are determined from the target supervision business request or the multi-source supervision business data; The construction documents to be reviewed are subjected to chapter identification and content element extraction to obtain the structured content of the construction documents. Based on the structured content of the construction data, determine the compliance review items, completeness review items, matching degree review items, and risk logic review items; Based on the compliance review items, the integrity review items, the matching degree review items, and the risk logic review items, the corresponding review basis is retrieved from the supervision business knowledge index to obtain the target review basis; Based on the aforementioned target review criteria, the structured content of the construction data is subjected to standard clause matching, key module missing detection, construction content consistency comparison, and risk logic verification to obtain problem identification results. Based on the problem identification results and the target review criteria, a structured review result containing problem points, supporting content, and rectification suggestions is generated, and the structured review result is used as the initial processing result.

8. The intelligent processing method for power engineering supervision business based on a large model according to claim 1, characterized in that, When the business processing type is risk verification, the step of retrieving the target business object, target object association, and target source identifier that match the target supervision business request from the supervision business knowledge index according to the business processing type and the processing strategy, and generating a business context based on the target business object, the target object association, and the target source identifier, includes: Acquire on-site identification data collected by on-site acquisition equipment; The on-site identification data is used to identify objects to obtain on-site objects to be verified, including personnel to be verified. The individuals to be verified are matched with the baseline information of the individuals to obtain the identity matching result and the identity matching confidence level. If the identity matching confidence level meets the preset identity confirmation conditions, the corresponding personnel information, entry approval information, qualification certificate information, examination record information and violation record information are retrieved from the supervision business knowledge index according to the identity matching result to obtain the on-site object association data; Based on the on-site object association data, determine the identity status, application status, qualification status, examination status, and historical risk status of the personnel to be verified; The identity status, the application status, the qualification status, the examination status, and the historical risk status are matched with preset risk verification rules to obtain the risk hit result; Based on the risk hit results, a risk verification result is generated, and the risk verification result is used as the initial processing result.

9. A power engineering supervision business intelligent processing system based on a large model, used to execute the method described in any one of claims 1 to 8, characterized in that, The system is deployed on the supervision business management platform and / or the supervision mobile application terminal, and the system includes: The data acquisition module is used to acquire the target supervision business request and multi-source supervision business data related to the target supervision business request. The multi-source supervision business data includes at least one of project data, on-site record data, personnel information data, quality and safety data, and management process data. The parsing and extraction module is used to perform format parsing and semantic extraction on the multi-source supervision business data to obtain business objects, object association relationships between the business objects, and source identifiers used to characterize the source of the business objects; An index building module is used to build a supervision business knowledge index based on the business object, the object association relationship, and the source identifier. The supervision business knowledge index includes a vector index, a keyword index, and a structured association index. The intent recognition module is used to recognize the intent of the target supervision business request and determine the business processing type and processing strategy corresponding to the target supervision business request. The business processing type includes at least one of information query, data analysis, document generation, document review and risk verification. The context generation module is used to retrieve, according to the business processing type and the processing strategy, the target business object, the target object association relationship and the target source identifier that match the target supervision business request from the supervision business knowledge index, and generate a business context based on the target business object, the target object association relationship and the target source identifier; The large model processing module is used to call the corresponding prompt word template and / or business processing rules according to the business processing type, input the target supervision business request and the business context into the large model, and generate the initial processing result; The verification and tracing module is used to perform consistency verification and tracing annotation on the initial processing result based on the processing strategy, the target object association relationship and the target source identifier, so as to obtain the target processing result; The result output module is used to output the target processing result; Specifically, the index building module is used for: Based on the object attributes of the business object, the business object is classified into business categories to obtain at least one of the following categories: project object, document object, personnel object, quality and safety object, process object, and standard basis object. The classification objects and the standardized business content corresponding to the classification objects are vectorized to obtain vector indexes; Keyword index is obtained by extracting keywords and generating tags from the classification objects and their corresponding object attributes. Based on the object association relationships, a structured association index is generated to represent the business association paths between different categories of objects; The source identifier is mapped to the vector index, the keyword index, and the structured association index respectively to obtain the supervision business knowledge index.

Images