CN-122021602-A - Environmental report generation method and question-answer report integrated system based on large model

CN122021602ACN 122021602 ACN122021602 ACN 122021602ACN-122021602-A

Abstract

The invention belongs to the technical field of data processing. The environment report generating method based on the large model and the question-answer report integrated system are provided, user requirements containing information such as report types, monitoring time periods and the like are received and packaged into structural configuration information, a main agent checking parameter generates task ID, task attributes are obtained through analysis, MCP tool resources are preallocated, the tasks are divided into subtasks such as data cleaning and the like, the subtasks are clearly dependent on the corresponding relation with the tools, the subtasks are distributed to corresponding sub agent clusters according to cooperative similarity, each sub agent invokes the resources to process multi-source data and generate charts, forms and professional conclusions, and after the matching degree is verified with the main intelligent experience, the complete environment report is integrated and generated. The invention improves the report generation efficiency and the professionality, ensures the report standardization and adapts to various environmental monitoring analysis requirements.

Inventors

GAO JIE
LI HUIJUAN
CHEN LINLIN
ZHANG CHAO
LIU XIAOFEI
WANG XINGLONG

Assignees

山东亿云信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (10)

1. A method for generating an environmental report based on a large model, comprising the following steps: Acquiring a report demand of a user, wherein the report demand comprises configuration information of report types, monitoring time periods, target areas, key pollutants and analysis dimensions, and packaging the configuration information into structural configuration information; The main agent receives the structured configuration information, generates a unique task ID after checking parameter integrity, analyzes report types, data scales and analysis complexity in the structured configuration information to obtain task attributes, wherein the task attributes comprise task priority, waiting time and required MCP tool types, and pre-distributes MCP tool resources according to the task attributes; The main agent decomposes the report generating task into subtasks of data cleaning, chart generation, table generation, professional analysis and format calibration based on the structural configuration information, task attributes and pre-allocated MCP tool resources, and the corresponding relation between each subtask and the pre-allocated MCP tool resources and the dependency relation between the subtasks are defined; the main agent calculates the cooperative similarity of the sub-agents and each sub-task and pre-distributing MCP tool resources, distributes the sub-tasks to a sub-agent cluster according to the cooperative similarity and associates task IDs, wherein the sub-agent cluster comprises a data management agent, a statistical chart agent, a form agent, a professional analysis agent and a format verification agent; Each sub-agent calls pre-allocated MCP tool resources according to the corresponding relation, calls multi-source data according to the dependency relation to execute processing, temporarily stores the processed data according to task ID, and synchronously generates a professional chart, a structured form and pollution tracing and trend prediction conclusion; and the format verification agent verifies the generated content of each sub-agent according to the constraint of the preset template, and the main agent verifies the matching degree of the verification content and the preset template, and then the main agent integrates the verification content and the preset template to generate a complete report after reaching the standard.
2. The method for generating a large model-based environmental report according to claim 1, The report requirement comprises a custom report requirement and a common report requirement, the configuration information of the custom report requirement also comprises a chapter frame for supporting a user to adjust the sequence of the chapter frame or supplement special analysis requirements, and the configuration information of the common report requirement also comprises a result receiving mode which comprises streaming real-time receiving and task progress tracking.
3. The method for generating a large model-based environmental report according to claim 1, The required MCP tool types comprise data calculation MCP, picture processing MCP and vectorization MilvusMCP, and the main agent determines the pre-allocation sequence of MCP tool resources, comprising: ; Wherein, the The task priority determined for the primary agent, 、 And The method comprises the steps that weight parameters calibrated through historical tasks are used for a main intelligent agent; A score representing the pre-allocation order, A normalized value representing the waiting time is provided, Representative of The type of tool to be used is a type of tool, Representative time; A normalized value representing the resource demand.
4. The method for generating a large model-based environmental report according to claim 1, The corresponding relation is that the subtask corresponding data of data cleaning calculates MCP, the subtask corresponding picture of the chart generation processes MCP, the subtask corresponding vectorization MilvusMCP of professional analysis, the subtask generated by the form does not relate to MCP tool resources with the subtask of format calibration; The dependency relationship is that after the subtasks of data cleaning are completed, the subtasks of chart generation and form generation are synchronously executed, after the subtasks of form generation are completed, the subtasks of professional analysis are executed, and after the subtasks of professional analysis are completed, the subtasks of format calibration are executed.
5. The method for generating a large model-based environmental report according to claim 1, Main agent verification score based on consistency Time of task completion Normalized value of (5), user satisfaction MCP tool calling efficiency construction rewarding function Comprising: ; In the formula, The efficiency is invoked for the MCP tool, MCP efficiency weights calibrated for the primary agent; representing a consistency score weight; Representing task time weights; Representing a user satisfaction weight.
6. A large model-based environmental report generation system, comprising: The demand configuration unit is configured to acquire the report demand of a user, wherein the report demand comprises configuration information of report types, monitoring time periods, target areas, key pollutants and analysis dimensions, and the configuration information is packaged into structural configuration information; The task initialization unit is configured to receive the structured configuration information, generate a unique task ID after checking parameter integrity, analyze report type, data scale and analysis complexity in the structured configuration information to obtain task attributes, wherein the task attributes comprise task priority, waiting time and required MCP tool type, and pre-allocate MCP tool resources according to the task attributes; The task decomposition unit is configured to decompose the report generation task into subtasks of data cleaning, chart generation, form generation, professional analysis and format calibration based on the structural configuration information, the task attribute and the pre-allocated MCP tool resources by the main agent, and define the corresponding relationship between each subtask and the pre-allocated MCP tool resources and the dependency relationship between the subtasks; The intelligent distribution unit is configured to calculate the cooperative similarity of the sub-intelligent agent and each sub-task and pre-distributing MCP tool resources by the main intelligent agent, distribute the sub-tasks to a sub-intelligent agent cluster according to the cooperative similarity and associate task IDs, wherein the sub-intelligent agent cluster comprises a data management intelligent agent, a statistical chart intelligent agent, a form intelligent agent, a professional analysis intelligent agent and a format verification intelligent agent; The subtask execution unit is configured to call pre-allocated MCP tool resources according to the corresponding relation by each sub-agent, call multi-source data according to the dependency relation to execute processing, temporarily store the processed data according to the task ID, and synchronously generate a professional chart, a structured form and pollution tracing and trend prediction conclusion; and the report integrating unit is configured to verify the generated content of each sub-agent by the format verification agent according to the constraint of the preset template, and integrate the verification content with the preset template to generate a complete report after the matching degree of the verification content and the preset template is verified by the main agent.
7. A question-answer report integrated system based on a large model is characterized in that, An intelligent report generation module and an intelligent question-answering module, the intelligent report generation module configured to perform the process of the big model based environmental report generation method of any one of claims 1-5; the intelligent question and answer module is configured to execute the following processes: acquiring multi-mode input data and function configuration information of a user, and establishing a tree-shaped dialogue node after analysis; based on the information recorded by the tree dialogue nodes, extracting corresponding multi-mode data and carrying out standardization processing; Extracting feature vectors of all modes after standardization, calculating dynamic weights through semantic similarity matching, and generating global feature vectors through weighting and fusion of a multi-head attention mechanism; Based on the global feature vector recognition mechanism factors, combining question and answer result errors and user feedback, and adopting reinforcement learning optimization mechanism factor parameters; and generating a professional optimization prompt word by combining the global feature vector, the environment parameters associated with the tree dialogue node and the optimized mechanism factor parameters, and generating a question-answering result based on the professional optimization prompt word.
8. The large model based question-answer report integration system of claim 7 in which, The multi-mode input data comprises atmospheric environment professional problems in text form, pollutant related terms of voice transcription, pollution image data and pollutant time sequence monitoring data, and the functional configuration information comprises calling an atmospheric professional knowledge base, enabling deep reasoning and starting networking supplementary data.
9. The large model based question-answer report integration system of claim 7 in which, The creating rule of the tree dialogue node is that a root node is generated when a new dialogue is initiated, input data, function configuration information and creating time are recorded, when a user triggers a retry operation, an original node is positioned according to a session identifier associated with the tree dialogue node, a child node is created by taking the original node as a father node, and the input data after the retry and the new function configuration information are synchronously recorded.
10. The large model based question-answer report integration system of claim 7 in which, Weighted fusion to generate global feature vector via multi-head attention mechanism Comprising: ; In the formula, , , Attention weights of text, image and time sequence modes respectively satisfy ; Representing a text feature vector; representing an image feature vector; Representing a timing feature vector; representing the global feature vector.

Description

Environmental report generation method and question-answer report integrated system based on large model Technical Field The invention relates to the technical field of data processing, in particular to an environment report generation method based on a large model and a question-answer report integrated system. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. Along with the continuous evolution of artificial intelligence technology, the technology such as a natural language large model, multi-mode data fusion, multi-agent cooperation and the like has shown remarkable value in the intelligent upgrading of various industries, and especially in the professional field of atmospheric environment monitoring and analysis, the technology fusion application becomes a core power for promoting the efficient development of the industries. The atmospheric environment monitoring work needs to integrate multidimensional information such as pollutant real-time monitoring and meteorological conditions, so that an analysis report with professionality and standardization is provided for environmental protection supervision and scientific research decision, and real-time interaction requirements of users in the scenes such as data exploration, question answering and the like are met. The maturation of natural language processing technology provides possibility for breaking the traditional manual processing mode, and the development of technologies such as multi-agent collaborative architecture, stream communication, tree data management and the like further lays a foundation for constructing an intelligent tool. At present, the requirements of an integrated system integrating intelligent report generation and intelligent question-answering functions in the industry are increasingly highlighted, by means of semantic understanding and generation capacity of a natural language large model and combining professional field knowledge, efficient processing, standard presentation and real-time interaction of multi-source data are realized, the integrated system becomes an important direction of technical upgrading in the field of atmospheric environment monitoring, and power-assisted related mechanisms improve decision-making efficiency and analysis depth. In the intelligent practice in the field of atmospheric environment monitoring, the prior art still has obvious short plates, and is difficult to meet the deep requirements of professional scenes. On the one hand, the report generation related technology lacks a systematic cooperative mechanism, cannot realize scientific decomposition of tasks and efficient cooperation of multiple intelligent agents, has insufficient integration capability on multi-source heterogeneous data, lacks an effective template constraint and accurate tool matching mechanism, causes insufficient normalization of a report generation process, is difficult to adapt to customized analysis requirements, and cannot output an analysis report with both professionality and consistency efficiently. On the other hand, the intelligent question-answering related technology has obvious limitation, lacks unified processing and deep fusion capability on multi-mode data such as texts, images and the like, lacks a structured dialogue management mode, is difficult to ensure logic continuity and traceability in complex interaction, lacks professional depth of semantic understanding, lacks a dynamic optimization mechanism, and cannot continuously improve accuracy and pertinence of question-answering according to user requirements and feedback. The two problems are related to each other, so that the comprehensive efficiency of an intelligent system in the field of atmospheric environment monitoring is restricted, integrated and high-quality intelligent service cannot be provided for users, and breakthrough is needed to be made through technical innovation. Disclosure of Invention In order to solve the defects of the prior art, the invention provides a large-model-based environment report generation method and a question-answer report integrated system, which solve the problems of difficult integration of multi-source data, low processing efficiency of complex tasks, poor format consistency and blindness of tool call in the traditional report generation process, overcome the defects of complicated manual operation, unsmooth subtasks cooperation and insufficient professional analysis depth in the prior art, improve the intelligent level and professional fitness of report generation, improve the cooperative efficiency of multi-source data processing and the decomposition execution efficiency of complex tasks, and enhance the normalization of report formats and the reliability of contents. In order to achieve the above purpose, the present invention adopts the following technical scheme: In a first aspect, the present inven