Search

CN-121996695-A - Intelligent analysis method and system for safety accident report based on multi-agent cooperation

CN121996695ACN 121996695 ACN121996695 ACN 121996695ACN-121996695-A

Abstract

The invention discloses a security accident report intelligent analysis method and system based on multi-agent cooperation, wherein the method comprises the following steps of analyzing a security accident investigation report, extracting key information to generate structured data, and storing the structured data in a database; the method comprises the steps of carrying out semantic search on a database by utilizing a large language model based on a query instruction, analyzing the searched data to generate an analysis report, extracting data which can be used for visualization from the analysis report, grouping the data, outputting grouping types, purposes and applicable chart types of the data to match corresponding chart templates from a chart template type library, and inputting the query instruction, the statistical data and the chart templates into the large language model to output a visualized chart. According to the invention, through the cooperative work of multiple agents, the full-flow automatic processing from multi-mode analysis, intelligent extraction, deep analysis to professional visualization of the accident report is realized, and the report analysis efficiency and decision support capability are obviously improved.

Inventors

  • QI HONGTUO
  • SHAN WENCHEN
  • QU HAIXIANG
  • LIU JIEPENG

Assignees

  • 重庆大学

Dates

Publication Date
20260508
Application Date
20251222

Claims (10)

  1. 1. The intelligent analysis method for the security accident report based on the multi-agent cooperation is characterized by comprising the following steps of: s1, collecting a security accident investigation report, analyzing the report by adopting a multi-mode information identification technology, extracting key information from the report to generate structural data, and storing the structural data into a database; S2, receiving a user query instruction, and carrying out semantic retrieval on a database by using a large language model based on the query instruction; s3, extracting data available for visualization from the analysis report, grouping the data based on the multidimensional characteristics of the data and the applicable chart types, and outputting the grouping category, the application and the applicable chart types of each group of data; S4, matching corresponding chart templates from a chart template type library by utilizing a retrieval enhancement generation technology based on the grouping category, the application and the applicable chart types of the data; s5, inputting the query instruction, the visual data and the chart template information into a large language model to generate a visual chart code, calling a visual rendering engine, and outputting the generated visual chart code into a visual chart.
  2. 2. The intelligent analysis method for security incident reports based on multi-agent collaboration according to claim 1, wherein in step S1), the database is established by the steps of: S1.1, collecting a security accident investigation report, and analyzing and extracting text texts, tables and images in the report by adopting a multi-mode information identification technology; S1.2, searching relevant information in the content extracted in the step S1.1) based on a preset system prompt word, and marking the relevant information as a null value if the relevant information related to the preset system prompt word is not found or the expression of the relevant information is not clear in the extracted content; The prompt words comprise report file numbers, report names, area information, time information, project names, project types, related affair unit information, accident types, lethal reasons, accident property identification conclusions, casualty personnel information, accident description, direct reason analysis and indirect reason analysis; S1.3, integrating key information corresponding to each prompting word to generate a structured text, wherein the structured text comprises prompting words and key information corresponding to the prompting words; s1.4, respectively storing the structured text and the safety accident investigation report into a database; the database comprises two sub-databases, and the structured text and the security incident investigation report are respectively stored in the two sub-databases.
  3. 3. The intelligent analysis method for security incident reports based on multi-agent collaboration according to claim 1, wherein in step S2), the method for generating the analysis report comprises the following steps: S2.1, based on a query instruction of a user, mobilizing a large language model to perform semantic retrieval on a database to acquire related key information; S2.2, counting the total number of safety accident investigation reports in the existing database; Calling out security accident investigation reports related to related key information from the existing database as related reports; The related report at least relates to one piece of related key information; S2.3, performing multidimensional depth analysis on the related report to generate an analysis report; the multi-dimensional depth analysis includes analyzing at least one dimension of a direct cause, an indirect cause, a potential problem, a motivation and intent of personnel, management responsibility, an organizational cultural factor, and a systematic defect of an incident; The analysis report comprises abstract and summary, accident key information and information carding, accident cause analysis, accident data analysis and analysis conclusion and suggestion; S2.4, calling out security incident investigation reports covered by the analysis report as coverage reports, counting the number of the coverage reports, and generating a security incident investigation report list based on the coverage reports to be used as an attached page of the analysis report; before generating the safety accident investigation report list, the number of the coverage reports needs to be checked in a database, and the coverage reports need to cover all relevant key information.
  4. 4. The intelligent analysis method of the security incident report based on multi-agent cooperation according to claim 1, wherein in step S3), the data is extracted according to the data which is explicitly given in the analysis report, is complete and can be compared and counted according to time, space or category; the multidimensional features include report file number, report name, region information, time information, project name, project type, incident unit information, incident type, cause of death, incident property identification conclusion, casualty information, incident pass description, direct cause analysis, and indirect cause analysis.
  5. 5. The intelligent analysis method for the security incident report based on multi-agent cooperation according to claim 1, wherein in step S4), a plurality of chart templates and scene descriptions applicable to the chart templates are pre-stored in the chart template type library; Each chart template is calibrated with applicable data types and typical purposes; In the matching process, the data category and the purpose are used as search conditions, vector similarity matching is carried out in a graph template type library, and one or more graph templates with highest matching degree are selected from the graph templates to be used as optimal graph templates.
  6. 6. The intelligent analysis method for security incident report based on multi-agent collaboration according to claim 1, wherein in step S5), the chart template information comprises an optimal chart template, a typical use of the optimal chart template, and ECharts codes of the optimal chart template.
  7. 7. The system for applying the intelligent analysis method for the security incident report based on multi-agent cooperation according to any one of claims 1-6 is characterized by comprising a survey report analysis module, an information intelligent extraction module, a structured information storage module, an intelligent analysis report generation module and a chart intelligent generation module; The investigation report analysis module is used for analyzing the security accident investigation report through a multi-mode information identification technology and analyzing text, form and image information in the report; The information intelligent extraction module extracts key information from the analyzed information based on a preset system prompt word; The system comprises a structured information storage module, a semantic similarity retrieval module and a database management module, wherein the structured information storage module is used for carrying out coding storage on key information and maintaining the state of the database through a session variable mechanism; The intelligent analysis report generation module is used for responding to the inquiry of a user, retrieving related data from the structural information storage module, summarizing to obtain the total number N of reports and the related report number M, driving the large language model to carry out multidimensional deep analysis to generate an analysis report, and counting the number K of covered cases by analysis by the summarized data in the data analysis and analysis links. Meanwhile, a data integrity checking function is integrated, and the consistency of the total number N of reports, the number M of related reports and the number K of statistical coverage reports is automatically verified before the report is output; The chart intelligent generation module is used for intelligently extracting relevant statistical data according to the generated analysis report and generating a visual chart by combining the matched chart template.
  8. 8. The system for applying the intelligent analysis method of the security incident report based on the multi-agent cooperation according to claim 7, wherein the investigation report analysis module is integrated with an optical character recognition unit, a table structure recognition unit and an image content analysis unit, can support the input of a plurality of document formats, and automatically recognizes and processes the incident investigation report templates of different areas and departments; The optical character recognition unit is used for reading texts, tables and characters in images in the report, the table structure recognition unit is used for recognizing and reconstructing the table structure in the report, and the image content analysis unit is used for extracting live photos and schematic diagrams in the report.
  9. 9. The system for applying a multi-agent collaboration-based security incident report intelligent analysis method as claimed in claim 7, wherein the chart intelligent generation module comprises a retrieval enhancement generation engine, a code generator and the rendering engine; the search enhancement generation engine is used for carrying out similarity matching in the chart template type library and screening out an optimal chart template from the chart template type library; the code generator is used for generating executable code conforming to ECharts specifications; The rendering engine is used to execute the code and output a final graph.
  10. 10. The system for applying the intelligent analysis method of the security incident report based on the multi-agent cooperation according to claim 7, wherein the system adopts a multi-agent cooperation mechanism, and each agent performs asynchronous communication and cooperation through a message or task queue to realize parallel processing and result cross-validation of analysis tasks; The intelligent agent comprises a text analysis intelligent agent, a causal reasoning intelligent agent, a statistical analysis intelligent agent, a visual chart generation intelligent agent and a strategy evaluation intelligent agent; the text analysis agent is used for analyzing the report text; The causal reasoning intelligent agent is used for analyzing and reasoning the accident cause; The statistical analysis agent is used for extracting statistical data suitable for visual display from the analysis report; the visualized chart generating agent is used for screening chart templates and visually presenting statistical data on the chart templates; The policy evaluation agent is used to review and optimize conclusions and suggestions in the analysis report.

Description

Intelligent analysis method and system for safety accident report based on multi-agent cooperation Technical Field The invention relates to the technical field of production safety supervision and management and intelligent information processing, in particular to an intelligent analysis method and system for a safety accident report based on multi-agent cooperation. Background The production safety accident investigation report is an important basis for deeply analyzing the safety production situation and summarizing the safety accident rule, and can provide reliable information support and scientific decision basis for the safety production work. However, the number of production incident survey reports tends to be very large, and the regulatory authorities encounter challenges in making statistics and analysis. Currently, statistics and analysis of accident investigation reports mainly depend on manual processing, and a supervision department needs to review a large number of investigation reports to extract key information. On the one hand, the way of relying on manual statistics alone is quite inefficient. The investigation reports are various in hierarchy and different in format, most of investigation reports are low in information density, and a large amount of manpower and material resources are consumed for manual statistics and analysis. On the other hand, the manual statistics mode is difficult to deeply mine information of a large number of reports, and is difficult to find deep reasons and commonalities of similar accidents, so that resource waste is caused. In addition, the manual statistics mode is strong in subjectivity, and an analyst easily ignores or excessively pays attention to some information to influence the accuracy and objectivity of analysis. In recent years, a large language model shows strong capability in terms of natural language processing, but when complex industry reports are processed, the problems of incomplete information extraction, opaque analysis logic, non-professional chart generation and the like still exist. Meanwhile, although the low-code platform reduces the application development threshold, how to deeply integrate the low-code platform with the AI capability and construct an end-to-end industrial solution, in particular to realize an automatic flow from unstructured reporting to deep analysis insight and even professional visual charts, and the technical problem which is not solved well in the prior art is still solved. Disclosure of Invention The invention aims to provide an intelligent analysis method for a security accident report based on multi-agent cooperation, which comprises the following steps: s1, collecting a security accident investigation report, analyzing the report by adopting a multi-mode information identification technology, extracting key information from the report to generate structural data, and storing the structural data into a database; S2, receiving a user query instruction, and carrying out semantic retrieval on a database by using a large language model based on the query instruction; s3, extracting data available for visualization from the analysis report, grouping the data based on the multidimensional characteristics of the data and the applicable chart types, and outputting the grouping category, the application and the applicable chart types of each group of data; S4, matching corresponding chart templates from a chart template type library by utilizing a retrieval enhancement generation technology based on the grouping category, the application and the applicable chart types of the data; s5, inputting the query instruction, the visual data and the chart template information into a large language model to generate a visual chart code, calling a visual rendering engine, and outputting the generated visual chart code into a visual chart. Further, in step S1), the establishment of the database comprises the steps of: S1.1, collecting a security accident investigation report, and analyzing and extracting text texts, tables and images in the report by adopting a multi-mode information identification technology; S1.2, searching relevant information in the content extracted in the step S1.1) based on a preset system prompt word, and marking the relevant information as a null value if the relevant information related to the preset system prompt word is not found or the expression of the relevant information is not clear in the extracted content; The prompt words comprise report file numbers, report names, area information, time information, project names, project types, related affair unit information, accident types, lethal reasons, accident property identification conclusions, casualty personnel information, accident description, direct reason analysis and indirect reason analysis; S1.3, integrating key information corresponding to each prompting word to generate a structured text, wherein the structured text comprises prompting wor