An Automatic Review Method and System for Environmental Monitoring Reports Based on Multi-Agent and Professional Retrieval Enhancement

Abstract

This invention relates to the field of natural language processing technology and discloses an automatic review method and system for environmental monitoring reports based on multi-agent and professional retrieval enhancement. Addressing the problems of low efficiency in manual review of environmental monitoring reports, difficulty in correlating and verifying multi-source data, and frequent updates to environmental standards, this invention acquires multi-source monitoring documents and extracts the original text through optical character recognition. After semantic parsing to extract core fields, it generates standard structured data using sample numbers as identifiers. The review task is then broken down and dispatched to agents for data consistency, standard compliance, logical relationships, and integrity for parallel verification. The latest standard knowledge base is dynamically invoked through a retrieval-enhanced generation (RAG) mechanism for compliance verification. Finally, the judgment results are summarized and conflict verification is performed to generate standardized review results containing problem location and standard basis. This achieves fully automated review of complex logic across documents, improving review efficiency, accuracy, and dynamic standard adaptation capabilities.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to an automatic review method and system for environmental monitoring reports based on multi-agent and professional retrieval enhancement. Background Technology

[0002] Environmental monitoring reports are the core basis for environmental law enforcement, discharge permit review, and environmental quality assessment. Their accuracy and compliance directly affect the scientific nature of ecological and environmental protection decisions. However, in practice, a complete environmental monitoring report typically consists of multiple independent documents, including the official report, the engagement contract, and original on-site and laboratory records, often stored in unstructured or semi-structured PDF format. This makes the review process extremely cumbersome and requires not only cross-document comparison of massive amounts of data but also strict adherence to complex and frequently updated environmental standards, resulting in a highly specialized and complex process.

[0003] Currently, the review of such reports mainly relies on manual verification, automated systems based on rule engines, or general-purpose large language model technology. However, these solutions all have significant limitations. Manual review is costly, inefficient, and susceptible to subjective factors; traditional rule engines lack flexibility, have extremely high maintenance costs, and cannot quickly adapt to frequent updates in environmental standards; while single large language models suffer from black-box problems when handling such professional reviews, are prone to "illusion" errors, and cannot accurately invoke the latest standard clauses, resulting in a lack of credibility and interpretability in the review results.

[0004] Furthermore, existing technologies generally lack deep multi-source data correlation capabilities, typically only allowing isolated verification of single reports. They cannot achieve cross-verification between reports, records, and contracts, making it difficult to identify hidden data inconsistencies or tampering issues. At the same time, the audit results output by various solutions vary in format, have vague problem identification, and lack clear standard criteria, making it impossible to directly connect with subsequent rectification or enforcement processes.

[0005] Therefore, the industry urgently needs an intelligent auditing mechanism that can deeply integrate multi-source data, dynamically retrieve the latest standards, and output standardized results. Summary of the Invention

[0006] This invention provides an automatic review method and system for environmental monitoring reports based on multi-agent and professional retrieval enhancement. This method replaces manual review, significantly improving review efficiency and reducing costs. By constructing a unified multi-source data structured standard, it achieves data integration and correlation verification of documents from different formats and sources. Utilizing a multi-agent collaboration mechanism, complex review tasks are broken down into specialized sub-tasks for parallel processing, effectively improving the accuracy and comprehensiveness of the review. Simultaneously, it introduces professional domain retrieval enhancement generation (RAG) technology to achieve dynamic management and precise retrieval of environmental standards, solving the problem of lagging standard updates. Ultimately, it establishes a standardized review result output system, enabling precise location of anomalies and comprehensively improving the traceability and credibility of review results.

[0007] This invention provides an automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement, comprising: S1. Obtain multiple environmental monitoring documents to be reviewed, perform optical character recognition on each environmental monitoring document to extract the original text data, and record the document name and page number of each environmental monitoring document to which the original text data belongs. S2. Based on the preset data field specifications, perform semantic parsing on the original text data, extract core fields including sample number, test items, corresponding standard number, time element, spatial element and process elements, and use the sample number as the association identifier to bind the core fields of the same sample number with its corresponding document name and page number to generate standard structured data. S3. Receive the audit task for the standard structured data and break it down into multiple audit sub-tasks. Send the multiple audit sub-tasks to the data consistency agent, standard compliance agent, logical relationship agent and integrity agent for parallel processing. S4. The standard compliance agent extracts the corresponding standard number and verifies the detection item using a pre-built standard knowledge base to obtain a compliance judgment result; the data consistency agent compares the values ​​of core fields with the same data item meaning under different document names to obtain a consistency judgment result; the logical relationship agent obtains a logical verification judgment result based on time and space elements; the integrity agent checks the missing process elements and core fields to obtain an integrity judgment result. S5. Summarize the judgment results of each agent. If there are contradictory conclusions among the judgment results of the same core field, trigger a second verification to obtain the final judgment result, and extract the document name and page number bound to the core field that failed the verification. Combine the corresponding judgment results to generate a standardized audit result.

[0008] Furthermore, S1 specifically includes: S101. Obtain multiple environmental monitoring documents awaiting review, which are input in batches and cover the entire process of environmental monitoring business; wherein, the environmental monitoring documents include formal environmental monitoring reports, testing entrustment contracts, original on-site sampling records, original laboratory analysis records, and qualification certificates; S102. Optical character recognition is performed on the scanned and editable versions of each of the environmental monitoring documents to extract the original text data, and the table row and column structure and paragraph hierarchy in each of the environmental monitoring documents are preserved. S103. Record the document name and page number of the environmental monitoring document to which each of the original text data belongs, and generate data information with coordinate position markers for each of the original text data, and establish a traceability record of text content-document name-page number-coordinate position.

[0009] Furthermore, S2 specifically includes: S201. Establish a preset data field specification containing core fields, and parse the original text data using a combination of rule matching and semantic parsing; wherein, the rule matching is used to identify the core fields with a preset fixed format, and the semantic parsing is used to perform contextual semantic understanding on unstructured text to extract the core fields. S202. Extract the core fields that conform to the preset data field specifications, and mark the document name and page number of the environmental monitoring document to which each core field belongs; S203. Establish an association index using the sample number as the association identifier, and bind the core fields belonging to different document names but having the same sample number, along with their corresponding document names and page numbers, to generate a standardized JSON data structure as the standard structured data; wherein, the association index is used to perform cross-document logical binding of the core fields scattered under different document names in the same batch.

[0010] Furthermore, S3 specifically includes: S301. Pre-set the data consistency agent, standard compliance agent, logical relationship agent, and integrity agent, and receive the overall audit task for the standard structured data; S302. Based on a preset scheduling mechanism, the overall audit task is automatically broken down into multiple audit sub-tasks with a single professional audit objective, and each audit sub-task is assigned to a corresponding intelligent agent. S303. Trigger each agent to process the audit sub-task in parallel and monitor the task execution status of each agent in real time.

[0011] Furthermore, S4 specifically includes: S401. The standard compliance intelligent agent extracts the corresponding standard number and verifies the detection item by combining it with the pre-built standard knowledge base to obtain a compliance judgment result. S402, The data consistency agent compares the values ​​of the core fields with the same data item meaning under different document names, and obtains the consistency judgment result by cross-comparing the numerical deviation of the same data item in each associated document; S403. The logical relationship intelligent agent performs logical rule verification based on the time element and the spatial element, and obtains the logical verification judgment result by verifying the time sequence of sampling, detection and report issuance, and verifying the consistency between the actual detection point and the contract agreement. S404. The integrity agent checks the missing information of the process requirements and the core fields, and obtains the integrity judgment result by checking the completeness of the required fields, legally required testing items, original record documents and personnel signature and seal requirements.

[0012] Furthermore, S401 specifically includes: After obtaining external standard documents and cleaning up pagination comments and page number noise, the documents are divided into blocks according to a hierarchical segmentation strategy and metadata containing standard number, file name, chapter identifier and sorting sequence number is attached to build a standard knowledge base. The standard knowledge base supports full reconstruction, incremental import, and single standard hot update, and performs usability verification including standard existence check and retrieval capability test after import. Inject a query tool bound to the standard knowledge base into the standard compliance intelligent agent, and normalize and perform variant expansion processing on the extracted corresponding standard numbers to construct search filtering conditions; Based on the search filtering conditions, semantic search is performed in the standard knowledge base to obtain candidate blocks, and the entire chapter is expanded and recalled for the hit chapter identifier. The target standard clauses are returned in order of the sorting sequence number. The recalled target standard clauses are injected into the audit context, the test item data is compared, and a list of issues including rule number, severity level, field location, evidence and standard basis is output to obtain the compliance judgment result.

[0013] Furthermore, S5 specifically includes: S501. Summarize the judgment results of each agent, compare the verification conclusions output by different agents for each core field, and if the conclusions for the same core field are mutually exclusive or logically contradictory, it is determined that there is an execution conflict. S502. For core fields where execution conflicts are detected, a secondary verification mechanism is triggered. By introducing a preset conflict resolution weight logic or performing higher-dimensional logic verification, the ambiguity of the judgment is eliminated, and the final judgment result set is output. S503. Based on the final judgment result set, filter out the core fields that failed the verification, extract the document name, page number and field name bound to them, and construct a standardized problem structure that includes problem type, problem location, problem description, standard basis and rectification suggestions; wherein, the problem location includes the document name, page number and coordinate position in the original text data of the environmental monitoring document to which the core field belongs. S504. Integrate the standardized issue structure into a unified output system, output an audit report and analyze the distribution indicators of each issue type, and generate an audit summary report including the audit pass rate; wherein, the output system supports generating at least one of the following: a visual audit report, an Excel-formatted issue list, and a structured JSON data file.

[0014] This invention also provides an automatic review system for environmental monitoring reports based on multi-agent and professional retrieval enhancement. Based on the aforementioned automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement, the system includes: The data input module is used to acquire multiple environmental monitoring documents to be reviewed, perform optical character recognition on each of the environmental monitoring documents to extract the original text data, and record the document name and page number of each environmental monitoring document to which the original text data belongs. The data structuring module is used to perform semantic parsing on the original text data based on preset data field specifications, extract core fields including sample number, test item, corresponding standard number, time element, spatial element and process element, and bind the core fields of the same sample number with its corresponding document name and page number to generate standard structured data. The audit task decomposition module is used to receive audit tasks for the standard structured data and decompose them into multiple audit sub-tasks, and send the multiple audit sub-tasks to the data consistency agent, standard compliance agent, logical relationship agent and integrity agent for parallel processing. The multi-agent audit module is used to extract the corresponding standard number through the standard compliance agent, and verify the detection item by combining it with a pre-built standard knowledge base to obtain a compliance judgment result; the data consistency agent compares the values ​​of core fields with the same data item meaning under different document names to obtain a consistency judgment result; the logical relationship agent obtains a logical verification judgment result based on time and space elements; and the integrity agent checks the missing process elements and core fields to obtain an integrity judgment result. The audit result generation module is used to summarize the judgment results of each agent. If there are contradictory conclusions among the judgment results of the same core field, a second verification is triggered to obtain the final judgment result. The document name and page number bound to the core field that failed the verification are extracted and a standardized audit result is generated in combination with the corresponding judgment result.

[0015] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0016] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0017] The beneficial effects of this invention are as follows: This invention firstly reduces the review time for a single report from several days to several hours through an automated mechanism that supports parallel processing, significantly improving overall review efficiency. Secondly, relying on multi-agent division of labor and cross-document data comparison technology, it improves the problem detection rate and ensures high consistency of output results for multiple repeated reviews, effectively solving the problems of missed detections and result fluctuations in manual review. In addition, this invention adopts a standard calling mechanism based on search-enhanced generation (RAG), which allows for rapid adaptation by simply adding data to the database when environmental standards are updated, without modifying system code, greatly reducing maintenance costs and enhancing standard adaptation capabilities. Finally, it can output structured review results containing precise problem locations and corresponding standard references, significantly improving the transparency, traceability, and interpretability of the review process. At the same time, with its flexible agent configuration and standard library content adjustment mechanism, it demonstrates excellent scalability to adapt to the review needs of various types of environmental monitoring reports. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the device structure according to an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.

[0021] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0023] This invention constructs an end-to-end closed-loop automatic review method based on a large vertical domain model. The whole consists of 5 steps, and each step communicates through a unified JSON data structure to complete the entire process of data input, parsing, structuring, review, and result output in sequence.

[0024] This invention primarily addresses the problems of low manual efficiency, inconsistent rule execution, difficulty in fully covering complex standards, and difficulty in standardizing results in the existing environmental monitoring report review process. Specifically, it aims to overcome the obstacle that unstructured PDFs and original records cannot be directly used for automated review, establish a unified correlation and verification mechanism between multi-source data (such as test reports and on-site and laboratory records), solve the pain point that environmental standards are complex and frequently updated, making it difficult to solidify them into static rules, break through the limitations of traditional single rule engines or single-model AI in dealing with complex review logic, and ultimately solve the defects of review results lacking unified expression and accurate problem location capabilities.

[0025] like Figure 1 As shown, this invention provides an automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement, including: S1. Obtain multiple environmental monitoring documents to be reviewed, perform optical character recognition on each environmental monitoring document to extract the original text data, and record the document name and page number of each environmental monitoring document to which the original text data belongs.

[0026] In one specific embodiment of the invention, step S1 includes the following sub-steps: S101. Obtain multiple environmental monitoring documents pending review.

[0027] Specifically, the system receives batches of environmental monitoring-related documents awaiting review. These documents are typically stored in unstructured or semi-structured PDF format. The acquired environmental monitoring documents comprehensively cover the entire environmental monitoring process, and their specific types include, but are not limited to: formal environmental monitoring reports, testing contracts, original on-site sampling records, original laboratory analysis records, qualification certificates, and other relevant supporting materials. Batch acquisition of these multi-source files provides a complete data foundation for subsequent cross-document automated review.

[0028] S102. Perform optical character recognition on each of the environmental monitoring documents to extract the original text data.

[0029] For the acquired scanned and editable documents, an Optical Character Recognition (OCR) interface is used to perform unified text recognition processing on each document. This extraction process not only converts the image or document content into editable text, but also simultaneously performs text extraction with location information. During the parsing process, while extracting the original text data of each document, the original feature information of each document is fully preserved, specifically including the table row and column structure and paragraph hierarchy within the document.

[0030] S103. Record the document name of the environmental monitoring document to which each of the original text data belongs and the page number in the environmental monitoring document.

[0031] While extracting the raw text data, the corresponding relationship between each extracted text segment and its physical source location is recorded in real time. Specifically, data information with coordinate location markers is generated for each extracted raw text data segment, and the document name of the environmental monitoring document to which the raw text data segment belongs, as well as the specific page number of the data segment in the corresponding document, are recorded simultaneously. By establishing a traceability record of "text content - document name - specific page number - coordinate location", traceable data support is provided for addressing cross-document comparisons in subsequent review steps, accurately locating review anomalies, and outputting standardized review criteria.

[0032] S2. Based on the preset data field specifications, perform semantic parsing on the original text data, extract core fields including sample number, test item, corresponding standard number, time element, spatial element and process element, and use the sample number as the association identifier to bind the core fields of the same sample number with its corresponding document name and page number to generate standard structured data.

[0033] In one specific embodiment of the invention, step S2 includes the following sub-steps: S201. Establish preset data field specifications and perform semantic parsing and rule matching on the original text data.

[0034] Specifically, a unified data field specification is predefined, comprehensively covering the core fields required for environmental monitoring operations. In addition to sample number, testing item, corresponding standard number (i.e., standard basis), time elements (specifically including sampling time and testing time), spatial elements (specifically including testing location), and process requirements, the core fields covered by this specification also include numerical values, units, and testing methods. When processing the raw text data extracted from documents of different formats, a combination of rule matching and semantic parsing is used for deep text understanding.

[0035] S202. Automatically extract the corresponding core fields and mark the source and location information of each core field.

[0036] Based on the results of the rule matching and semantic parsing described above, the corresponding core fields are automatically and accurately extracted from the unstructured raw text data. While extracting each core field, the document name and page number of the corresponding environmental monitoring document for each extracted field are strictly marked and recorded, thus ensuring that all extracted discrete fields have a clear tracing path.

[0037] S203. Construct an associated index using the sample number as the primary key, and perform cross-document data binding to generate a standardized JSON data structure.

[0038] After field extraction and information labeling are completed, a correlation index is established using the sample number as the association identifier (i.e., primary key). Based on this correlation index, the core fields of different related documents (such as formal reports, original records, etc.) with the same sample number but scattered within the same batch are cross-bound with their corresponding document names and page numbers, thereby achieving cross-document data association. Finally, the unstructured text data that has completed the association binding is converted into a standardized JSON data structure, i.e., generating the standard structured data for subsequent intelligent agents to call.

[0039] S3. Receive the audit task for the standard structured data and break it down into multiple audit sub-tasks. Send the multiple audit sub-tasks to the data consistency agent, standard compliance agent, logical relationship agent and integrity agent for parallel processing.

[0040] In one specific embodiment of the invention, step S3 includes the following sub-steps: S301. Set up a specialized intelligent agent and receive auditing tasks for the standard structured data.

[0041] Specifically, to address the complex and highly specialized review requirements of environmental monitoring reports, multiple specialized intelligent agents are pre-constructed. These specialized intelligent agents include: a data consistency agent, a standard compliance agent, a logical relationship agent, and an integrity agent. After completing cross-document data association and binding, the system receives the overall review task for the standard structured data generated in step S2.

[0042] S302. The audit task is automatically broken down into multiple audit sub-tasks and distributed to the corresponding specialized intelligent agents.

[0043] Specifically, upon receiving the audit task, the complex overall audit task is automatically broken down into multiple audit sub-tasks with a single professional audit objective based on an internal unified scheduling mechanism. Subsequently, according to the professional attributes of each sub-task, the multiple audit sub-tasks are accurately distributed to the corresponding data consistency agent, standard compliance agent, logical relationship agent, and integrity agent.

[0044] S303. Trigger each of the specialized intelligent agents to perform parallel processing and monitor the execution status in real time.

[0045] Specifically, upon receiving the corresponding review sub-task, each specialized intelligent agent synchronously triggers parallel review execution. During the parallel execution of the review sub-tasks by each specialized intelligent agent, the task execution status of each agent is monitored in real time to grasp the overall review progress and ensure the stability of multi-agent collaboration, thereby providing pre-operational support for the smooth aggregation and processing of all subsequent review results.

[0046] S4. The standard compliance agent extracts the corresponding standard number and verifies the detection item using a pre-built standard knowledge base to obtain a compliance judgment result; the data consistency agent compares the values ​​of core fields with the same data item meaning under different document names to obtain a consistency judgment result; the logical relationship agent obtains a logical verification judgment result based on time and space elements; and the integrity agent checks the missing process elements and core fields to obtain an integrity judgment result.

[0047] In one specific embodiment of the invention, step S4 includes the following sub-steps: S401. The standard compliance intelligent agent extracts the corresponding standard number and verifies the detection item by combining it with the pre-built standard knowledge base to obtain a compliance judgment result.

[0048] The execution of this step relies on the underlying RAG (Retrieval Enhancement Generation) mechanism, and specifically includes the following process: (1) Preparation for importing new standards and initialization of the knowledge base: Obtain newly released or revised external environment standard files (e.g., standard library files in Markdown format) and ensure that the file name or the beginning of the text has a recognizable standard number. When starting or executing the import script, initialize the standard knowledge base object, connect to the vector database (e.g., PostgreSQL with the PgVector plugin), and bind the preset embedding model; if a full rebuild is required, enable the recreate database cleanup function.

[0049] (2) Standard text cleaning, structuring, and metadata writing: When importing standard text, invalid text content such as pagination comments and page number noise is first cleaned. Then, a hierarchical segmentation strategy is used to divide the text into blocks to avoid the entire standard being entered into the database at once, which would lead to inaccurate subsequent retrieval. When each text block is written to the vector database, key metadata is forcibly attached. The key metadata includes: standard number (standard_code), file name (file_name), section identifier (section_id), and sorting sequence number (SSID), which serve as the underlying data foundation for accurate retrieval by standard number and extended retrieval by section.

[0050] (3) Vector Database Update Strategy and Availability Verification: The vector database supports three dynamic update methods: first, a full reconstruction after clearing and rebuilding (recreate=True); second, incremental import that only adds standards not yet included in the database and skips existing items (skip_if_exists); and third, single-standard hot update that first deletes old blocks by file name and then reloads the standard to solve the problem of standard revision coverage. After the import is completed, availability verification is performed immediately, which includes checking the existence of standards by standard number or file name, and testing the recall capability of new standards by entering clause questions.

[0051] (4) Audit Tool Injection and Standard Retrieval Initiation: In the overall audit process, the audit orchestrator injects a standard query tool (StandardsQueryTools) when initializing the verification agent. This tool is internally bound to the standard knowledge base and a custom retrieval tool, and enables the knowledge search function (search_knowledge). When the agent needs standard basis (such as sampling duration, detection limit, quality control requirements, etc.) for auditing a certain field, it calls the tool (query_standard_content) to initiate a retrieval. The tool layer automatically normalizes and performs variant expansion processing on the extracted corresponding standard numbers (for example, equivalently merging "GB / T" and "GB_T"), thereby constructing accurate knowledge filtering conditions (knowledge_filters).

[0052] (5) Semantic Retrieval, Enhanced Recall, and Compliance Determination: A custom retrieval tool extracts the standard number from the retrieval request and merges it with the knowledge filtering conditions. Semantic retrieval is performed in the vector database to obtain the Top-K candidate text blocks. If the hit results are concentrated in a specific section identifier (section_id), the entire section extended recall mechanism is triggered, and the complete content of the section is returned in order of the sorting sequence number (ssid). Finally, the standard compliance agent injects the recalled complete clauses into the audit context and compares the project data in the standard structured data item by item. If an anomaly is determined, a list of issues containing rule number, severity level, field location, evidence, and standard basis is output, ultimately forming a compliance determination result based on the latest standard.

[0053] S402. The data consistency agent compares the values ​​of core fields with the same data item meaning under different document names to obtain a consistency judgment result.

[0054] Specifically, the data consistency agent traverses all related documents, comparing the corresponding fields in the official test report with the original field sampling records, original laboratory analysis records, and the testing contract. Its core function is to verify the consistency of identical data items (such as the concentration of pollutants at a specific location) across documents, accurately marking anomalies such as inconsistent values ​​and information mismatches, and recording the specific details of the differences and their location in both documents. This effectively identifies data tampering or hidden inconsistencies, leading to a consistency determination result.

[0055] S403, the logical relation intelligent agent derives the logical verification and judgment result based on time and space elements.

[0056] Specifically, the logical relational agent is responsible for performing in-depth logical rule verification. Regarding time elements, it rigorously verifies the time logic of business processes, such as checking objective laws like "whether the sampling time is earlier than the detection time" and "whether the report issuance time is later than the detection time." Regarding spatial elements, it verifies whether the actual detection points are consistent with the points agreed upon in the contract. Furthermore, it verifies the data operation logic, such as checking whether the total data matches the sum of each component data precisely. For any item that violates objective logic and operation rules, it records logical anomalies and derives a logical verification judgment result.

[0057] S404. The integrity agent checks the missing elements and core fields of the process and obtains the integrity judgment result.

[0058] Specifically, the integrity agent performs a comprehensive audit. On one hand, it checks the completeness of the core data itself, verifying whether any required fields are missing, necessary legally mandated testing items are omitted, or previous original records are missing. On the other hand, it rigorously verifies the completeness of process requirements, focusing on whether signatures of relevant personnel and official seals are complete. If any items are missing, the type and location of the missing items are recorded in detail, and an integrity determination result is derived.

[0059] S5. Summarize the judgment results of each agent. If there are contradictory conclusions among the judgment results of the same core field, trigger a second verification to obtain the final judgment result, and extract the document name and page number bound to the core field that failed the verification. Combine the corresponding judgment results to generate a standardized audit result.

[0060] In one specific embodiment of the invention, step S5 includes the following sub-steps: S501. Summarize and monitor in real time the judgment results output by the data consistency agent, the standard compliance agent, the logical relationship agent, and the integrity agent. For each core field in the standard structured data, compare the verification conclusions given by different agents. If it is found that the conclusions output by different agents for the same core field are mutually exclusive or logically contradictory, it is determined that there is an execution conflict.

[0061] S502. For detected conflicting conclusions, a secondary verification mechanism is triggered. Under this mechanism, conflicting items are reviewed by introducing preset conflict resolution weight logic or performing higher-dimensional logic verification, thereby eliminating ambiguity and outputting a final, unified, and definitive review conclusion. If all agents reach the same conclusion or the secondary verification is completed, a final judgment result set for this batch of environmental monitoring documents is formed.

[0062] S503. Based on the final judgment result set, filter out the abnormal core fields that failed the validation. For each core field that failed the validation, extract the document name and corresponding page number information pre-bound in step S2, and accurately map its field name. Then, construct a standardized abnormal problem structure, which fully covers: problem type, specific problem location (including document name, page number, and field), detailed problem description, corresponding standard basis, and targeted rectification suggestions.

[0063] S504 integrates the generated standardized issue structure into a unified output system. In terms of output format, it supports generating visual audit reports, issue list Excel files, and structured JSON data files to adapt to different business processing needs. Simultaneously, it executes data statistics logic to calculate the overall pass rate of this audit, analyze the distribution indicators of each issue type, and finally generate an audit summary report, achieving accurate positioning of audit results, clear basis, and full process traceability.

[0064] like Figure 2 As shown, the present invention also provides an automatic review system for environmental monitoring reports based on multi-agent and professional retrieval enhancement. Based on the aforementioned automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement, the system includes: The data input module 1 is used to acquire multiple environmental monitoring documents to be reviewed, perform optical character recognition on each of the environmental monitoring documents to extract the original text data, and record the document name and page number of each environmental monitoring document to which the original text data belongs. Data structuring module 2 is used to perform semantic parsing on the original text data based on preset data field specifications, extract core fields including sample number, test item, corresponding standard number, time element, spatial element and process element, and bind the core fields of the same sample number with its corresponding document name and page number to generate standard structured data using the sample number as the association identifier. The audit task decomposition module 3 is used to receive the audit task for the standard structured data and decompose it into multiple audit sub-tasks, and send the multiple audit sub-tasks to the data consistency agent, standard compliance agent, logical relationship agent and integrity agent for parallel processing. The multi-agent audit module 4 is used to extract the corresponding standard number through the standard compliance agent, and verify the detection item by combining it with the pre-built standard knowledge base to obtain a compliance judgment result; the data consistency agent compares the values ​​of core fields with the same data item meaning under different document names to obtain a consistency judgment result; the logical relationship agent obtains a logical verification judgment result based on time and space elements; and the integrity agent checks the missing process elements and core fields to obtain an integrity judgment result. The audit result generation module 5 is used to summarize the judgment results of each agent. If there are contradictory conclusions among the judgment results of the same core field, a second verification is triggered to obtain the final judgment result. The document name and page number bound to the core field that failed the verification are extracted, and a standardized audit result is generated in combination with the corresponding judgment result.

[0065] Each of the above modules is used to perform the corresponding steps in the above-mentioned automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement. The specific implementation method is as described in the above method embodiments, and will not be repeated here.

[0066] like Figure 3 As shown, the present invention also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores all data required for the process of an automated review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement. The network interface is used for communication with external terminals via a network connection. The computer program is executed by the processor to implement the automated review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement.

[0067] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0068] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described automatic review methods for environmental monitoring reports based on multi-agent and professional retrieval enhancement.

[0069] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by hardware related to computer program instructions. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM), such as dynamic RAM (used as main storage) or static RAM (commonly used as cache memory). By way of illustration and not limitation, RAM has various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), and Rambus DRAM (RDRAM).

[0070] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0071] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement, characterized in that, include: S1. Obtain multiple environmental monitoring documents to be reviewed, perform optical character recognition on each of the environmental monitoring documents to extract the original text data, and record the document name and page number of each environmental monitoring document to which the original text data belongs. S2. Based on the preset data field specifications, perform semantic parsing on the original text data, extract core fields including sample number, test item, corresponding standard number, time element, spatial element and process element, and use the sample number as the association identifier to bind the core fields of the same sample number with its corresponding document name and page number to generate standard structured data. S3. Receive the audit task for the standard structured data and break it down into multiple audit sub-tasks. Send the multiple audit sub-tasks to the data consistency agent, standard compliance agent, logical relationship agent and integrity agent for parallel processing. S4. The standard compliance agent extracts the corresponding standard number and verifies the detection item by combining it with the pre-built standard knowledge base to obtain a compliance judgment result; the data consistency agent compares the values ​​of core fields with the same data item meaning under different document names to obtain a consistency judgment result; the logical relationship agent obtains a logical verification judgment result based on time elements and spatial elements. The integrity agent checks for missing process components and core fields to obtain an integrity determination result. S5. Summarize the judgment results of each agent. If there are contradictory conclusions among the judgment results of the same core field, trigger a second verification to obtain the final judgment result, and extract the document name and page number bound to the core field that failed the verification. Combine the corresponding judgment results to generate a standardized audit result.

2. The automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement as described in claim 1, characterized in that, S1 specifically includes: S101. Obtain multiple environmental monitoring documents awaiting review, which are input in batches and cover the entire process of environmental monitoring business; wherein, the environmental monitoring documents include formal environmental monitoring reports, testing entrustment contracts, original on-site sampling records, original laboratory analysis records, and qualification certificates; S102. Optical character recognition is performed on the scanned and editable versions of each of the environmental monitoring documents to extract the original text data, and the table row and column structure and paragraph hierarchy in each of the environmental monitoring documents are preserved. S103. Record the document name and page number of the environmental monitoring document to which each of the original text data belongs, and generate data information with coordinate position markers for each of the original text data, and establish a traceability record of text content-document name-page number-coordinate position.

3. The automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement as described in claim 1, characterized in that, S2 specifically includes: S201. Establish a preset data field specification containing core fields, and parse the original text data using a combination of rule matching and semantic parsing; wherein, the rule matching is used to identify the core fields with a preset fixed format, and the semantic parsing is used to perform contextual semantic understanding on unstructured text to extract the core fields. S202. Extract the core fields that conform to the preset data field specifications, and mark the document name and page number of the environmental monitoring document to which each core field belongs; S203. Establish an association index using the sample number as the association identifier, and bind the core fields belonging to different document names but having the same sample number, along with their corresponding document names and page numbers, to generate a standardized JSON data structure as the standard structured data; wherein, the association index is used to perform cross-document logical binding of the core fields scattered under different document names in the same batch.

4. The automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement as described in claim 1, characterized in that, S3 specifically includes: S301. Pre-set the data consistency agent, standard compliance agent, logical relationship agent, and integrity agent, and receive the overall audit task for the standard structured data; S302. Based on a preset scheduling mechanism, the overall audit task is automatically broken down into multiple audit sub-tasks with a single professional audit objective, and each audit sub-task is assigned to a corresponding intelligent agent. S303. Trigger each agent to process the audit sub-task in parallel and monitor the task execution status of each agent in real time.

5. The automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement as described in claim 1, characterized in that, S4 specifically includes: S401. The standard compliance intelligent agent extracts the corresponding standard number and verifies the detection item by combining it with the pre-built standard knowledge base to obtain a compliance judgment result. S402, The data consistency agent compares the values ​​of the core fields with the same data item meaning under different document names, and obtains the consistency judgment result by cross-comparing the numerical deviation of the same data item in each associated document; S403. The logical relationship intelligent agent performs logical rule verification based on the time element and the spatial element, and obtains the logical verification judgment result by verifying the time sequence of sampling, detection and report issuance, and verifying the consistency between the actual detection point and the contract agreement. S404. The integrity agent checks the missing information of the process requirements and the core fields, and obtains the integrity judgment result by checking the completeness of the required fields, legally required testing items, original record documents and personnel signature and seal requirements.

6. The automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement as described in claim 5, characterized in that, Specifically, S401 includes: After obtaining external standard documents and cleaning up pagination comments and page number noise, the documents are divided into blocks according to a hierarchical segmentation strategy and metadata containing standard number, file name, chapter identifier and sorting sequence number is attached to build a standard knowledge base. The standard knowledge base supports full reconstruction, incremental import, and single standard hot update, and performs usability verification including standard existence check and retrieval capability test after import. Inject a query tool bound to the standard knowledge base into the standard compliance intelligent agent, and normalize and perform variant expansion processing on the extracted corresponding standard numbers to construct search filtering conditions; Based on the search filtering conditions, semantic search is performed in the standard knowledge base to obtain candidate blocks, and the entire chapter is expanded and recalled for the hit chapter identifier. The target standard clauses are returned in order of the sorting sequence number. The recalled target standard clauses are injected into the audit context, the test item data is compared, and a list of issues including rule number, severity level, field location, evidence and standard basis is output to obtain the compliance judgment result.

7. The automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement as described in claim 1, characterized in that, S5 specifically includes: S501. Summarize the judgment results of each agent, compare the verification conclusions output by different agents for each core field, and if the conclusions for the same core field are mutually exclusive or logically contradictory, it is determined that there is an execution conflict. S502. For core fields where execution conflicts are detected, a secondary verification mechanism is triggered. By introducing a preset conflict resolution weight logic or performing higher-dimensional logic verification, the ambiguity of the judgment is eliminated, and the final judgment result set is output. S503. Based on the final judgment result set, filter out the core fields that failed the verification, extract the document name, page number and field name bound to them, and construct a standardized problem structure that includes problem type, problem location, problem description, standard basis and rectification suggestions; wherein, the problem location includes the document name, page number and coordinate position in the original text data of the environmental monitoring document to which the core field belongs. S504. Integrate the standardized issue structure into a unified output system, output an audit report and analyze the distribution indicators of each issue type, and generate an audit summary report including the audit pass rate; wherein, the output system supports generating at least one of the following: a visual audit report, an Excel-formatted issue list, and a structured JSON data file.

8. An automatic review system for environmental monitoring reports based on multi-agent and professional retrieval enhancement, based on the automatic review method for environmental monitoring reports based on multi-agent and professional retrieval enhancement as described in any one of claims 1 to 7, characterized in that, The system includes: The data input module is used to acquire multiple environmental monitoring documents to be reviewed, perform optical character recognition on each of the environmental monitoring documents to extract the original text data, and record the document name and page number of each environmental monitoring document to which the original text data belongs. The data structuring module is used to perform semantic parsing on the original text data based on preset data field specifications, extract core fields including sample number, test item, corresponding standard number, time element, spatial element and process element, and bind the core fields of the same sample number with its corresponding document name and page number to generate standard structured data. The audit task decomposition module is used to receive audit tasks for the standard structured data and decompose them into multiple audit sub-tasks, and send the multiple audit sub-tasks to the data consistency agent, standard compliance agent, logical relationship agent and integrity agent for parallel processing. The multi-agent audit module is used to extract the corresponding standard number through the standard compliance agent, and verify the detection item by combining it with a pre-built standard knowledge base to obtain a compliance judgment result; the data consistency agent compares the values ​​of core fields with the same data item meaning under different document names to obtain a consistency judgment result; the logical relationship agent obtains a logical verification judgment result based on time and space elements; and the integrity agent checks the missing process elements and core fields to obtain an integrity judgment result. The audit result generation module is used to summarize the judgment results of each agent. If there are contradictory conclusions among the judgment results of the same core field, a second verification is triggered to obtain the final judgment result. The document name and page number bound to the core field that failed the verification are extracted and a standardized audit result is generated in combination with the corresponding judgment result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images