A method and system for constructing an intelligent agent for intelligent auditing of a flood impact assessment report

Abstract

The application discloses a flood impact assessment report intelligent auditing intelligent agent construction method and system, the method comprises the following steps: after obtaining the input material, carrying out structured processing, generating a unified review knowledge unit; based on the candidate block boundary, the unified review knowledge unit is adaptively blocked to obtain a plurality of review blocks, the corresponding report original evidence and standard article basis are retrieved for each review block, and the evidence context is assembled; the ability portrait of the callable model is established, the task adaptation degree is calculated according to the ability demand weight of the review task and the model ability portrait, the model set meeting the preset condition is dynamically selected according to the adaptation degree, the evidence context is constrained, each review block is reviewed from multiple review perspectives, and each model review opinion is obtained. Dynamic weight fusion and conflict identification are carried out on the multiple model review opinions to generate fused review opinions; the fused review opinions of the multiple review blocks are cross-task consistency checked to generate the final auditing result.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a method and system for constructing an intelligent agent for the intelligent review of flood impact assessment reports. Background Technology

[0002] Flood impact assessment reports are crucial technical documents for river-related construction projects, flood control safety assessments, river management, and water administration approvals. Their content typically includes a project overview, basic information on rivers, lakes, and flood storage areas, flood control standards, hydrological analysis, impacts of backwater, scour, and levee / bank stability, impact assessment, measures to mitigate and eliminate impacts, conclusions, and recommendations. These reports involve significant technical judgment and must comply with relevant regulations, technical standards, guidelines, and review requirements.

[0003] Traditional review processes rely primarily on manual chapter-by-chapter reading and expert judgment, which suffers from problems such as long review cycles, complex sourcing of standards and references, inconsistent opinions among experts, and difficulty in timely detection of cross-chapter parameter contradictions. With the development of large language models and intelligent agent technology, automated report review using models has become possible. However, directly inputting the entire report into a single model for review is easily limited by factors such as context length, model expertise, accuracy of standard citations, and output stability, leading to missed reviews, misjudgments, or the inability to trace the review basis.

[0004] Therefore, a method for constructing an intelligent review agent for flood impact assessment reports is needed. This method should integrate report parsing, standard retrieval, review task decomposition, multi-model collaborative reasoning, fusion adjudication, and traceable output into a complete process, so that the review results are not only professional, but also stable, interpretable, and verifiable. Summary of the Invention

[0005] To address at least one deficiency or improvement need in the prior art, this invention provides a method and system for constructing an intelligent review agent for flood impact assessment reports. This solves the technical problems in the prior art, such as low review efficiency, complex standard reference search, difficulty in discovering cross-chapter parameter contradictions, and lack of stability and interpretability of review results, caused by reliance on manual review or a single model.

[0006] To achieve the above objectives, according to the first aspect of the present invention, a method for constructing an intelligent review agent for flood impact assessment reports is provided. This method involves acquiring input materials, including the flood impact assessment report to be reviewed, technical standard documents, regulatory guidelines, and supplementary review requirements; structuring the input materials to generate a unified review knowledge unit; adaptively dividing the unified review knowledge unit based on candidate block boundaries to obtain multiple review blocks; retrieving corresponding original report evidence and standard provisions for each review block and assembling them into an evidence context; establishing a capability profile of callable models; calculating task suitability based on the capability requirement weights of the review task and the model capability profile; dynamically selecting a set of models whose suitability meets preset conditions; and, under the constraints of the evidence context, independently reviewing each review block from multiple review perspectives to obtain review opinions from each model; dynamically weighting and conflict identification of the review opinions from multiple models to generate a fused review opinion; and performing cross-task consistency verification on the fused review opinions from multiple review blocks, including cross-chapter parameter consistency verification, problem merging, and risk ranking to generate the final review result.

[0007] In one exemplary embodiment, the unified review knowledge unit is represented as: ,in, Indicates the first i A unified knowledge unit for review Indicates a unique unit number. Represents text content, Indicates the content type. Indicates the source location. Indicates the chapter path, Indicates association rules, Represents a semantic vector.

[0008] In an exemplary embodiment, the adaptive segmentation of the unified review knowledge unit based on candidate segmentation boundaries includes: using chapter titles, paragraph boundaries, table boundaries, calculation explanation paragraphs, evaluation conclusion paragraphs, and measure description paragraphs as candidate segmentation boundaries, calculating the comprehensive segmentation cost for each candidate segmentation, expressed as: Cost(C' k )=αL(C' k )+βS(C' k )+γH(C' k ); where C' k This represents the k-th candidate block, Cost(C' k L(C') represents the block total cost of the k-th candidate block. k S(C') represents the cost of exceeding the length limit. k H(C') represents the semantic fragmentation cost. k) represents the cost of chapter-level segmentation, and α, β, and γ are weighting coefficients; with the goal of minimizing the overall cost of segmentation, the optimal segmentation scheme is determined from the candidate segments.

[0009] In an exemplary embodiment, the step of retrieving the corresponding original report evidence and standard provisions for each review block includes: combining the review block text and review matters into a review task; retrieving candidate evidence units related to the review task from the standard provisions knowledge unit set and the review knowledge unit set to form a candidate evidence set; calculating the retrieval score of each candidate evidence unit in the candidate evidence set based on the weighted sum of semantic relevance score, keyword matching score, and metadata matching score, and selecting candidate evidence units whose retrieval scores meet preset conditions as review basis; wherein, the metadata includes at least one of chapter path, content type, page number position, review topic tag, scope of application, or provisions type.

[0010] In an exemplary embodiment, the dynamic weighted fusion of the review opinions of the multiple models includes: normalizing the review opinions generated by each model and mapping the conclusion judgment to a set of candidate conclusions; calculating the fusion score of each candidate conclusion by combining the historical capability score of the model, the evidence coverage of the current output, the consistency score with the conclusions of other models, and the model's self-assessed confidence, and using the review opinion corresponding to the candidate conclusion with the highest fusion score as the fused review opinion.

[0011] In an exemplary embodiment, the method further includes: calculating the degree of conflict between the review opinions generated by different models during the process of dynamically weighting and fusing the review opinions of the multiple models; when the degree of conflict exceeds a preset threshold, re-searching the basis and conducting a second review, or marking the review task as pending manual review.

[0012] In one exemplary embodiment, the task adaptability is represented as: ,in, Indicates a callable model. This represents the k-th review block. This indicates the d-th review item. Represents a set of capabilities. This represents the weight of the current task's requirement for the r-th type of capability. Indicates a callable model The score on the r-th ability Indicates a callable model Response latency, These are the weighting coefficients.

[0013] According to a second aspect of the present invention, a system for constructing an intelligent review agent for flood impact assessment reports is also provided. This system executes the aforementioned method for constructing an intelligent review agent for flood impact assessment reports, comprising: acquiring input materials, including the flood impact assessment report to be reviewed, technical standard documents, regulatory guidelines, and supplementary review requirements; performing structured processing on the input materials to generate a unified review knowledge unit; adaptively dividing the unified review knowledge unit based on candidate block boundaries to obtain multiple review blocks; retrieving corresponding original report evidence and standard provisions for each review block and assembling them into an evidence context; establishing a capability profile of a callable model; calculating task suitability based on the capability requirement weights of the review task and the model capability profile; dynamically selecting a set of models whose suitability meets preset conditions; independently reviewing each review block from multiple review perspectives under the constraints of the evidence context to obtain review opinions from each model; dynamically weighting and conflict identification of the review opinions from the multiple models to generate a fused review opinion; and performing cross-task consistency verification on the fused review opinions from the multiple review blocks, including cross-chapter parameter consistency verification, problem merging, and risk ranking, to generate the final review result.

[0014] According to a third aspect of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer program, wherein the computer program is configured to execute the above-described method for constructing an intelligent auditing agent for flood impact assessment reports at runtime.

[0015] According to a fourth aspect of the present invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-described method for constructing an intelligent agent for reviewing flood impact assessment reports via the computer program.

[0016] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) This invention provides a method for constructing an intelligent review agent for flood impact assessment reports. Through an adaptive block-based approach, review blocks are generated by comprehensively considering chapter boundaries, semantic continuity, context length limitations, and the completeness of review items. This ensures that the blocks satisfy model context constraints while maintaining chapter completeness and semantic continuity as much as possible, avoiding context loss and information omissions caused by directly inputting long documents into the model. This ensures that the review covers every chapter and every review item of the report. Multiple models with different capabilities are invoked for collaborative review from multiple perspectives, including professional technology, standard compliance, formal logic, data consistency, and risk classification. The fusion adjudication process verifies the evidence constraints of the multiple model outputs and dynamically weights and fuses them, effectively reducing the risk of unstable review results or insufficient professional judgment from a single model, and improving the accuracy and reliability of the review conclusions.

[0017] (2) This invention imposes evidentiary constraints on each review opinion, requiring the simultaneous binding of evidence from the original report and the standard provisions, ensuring that each review opinion is verifiable and traceable. Reviewers can directly verify the source of the review conclusions, enhancing the credibility and reproducibility of the review results. By extracting key parameters, verifying cross-chapter consistency, and merging issues from multiple review segments, it can automatically identify parameter contradictions, inconsistencies in conclusions, and duplication issues that are difficult to identify in a single review segment, improving the systematicness and comprehensiveness of the review of long flood impact assessment reports.

[0018] (3) By constructing a complete intelligent review process, this invention automatically links report parsing, standard retrieval, review task decomposition, multi-model collaborative reasoning, fusion judgment and traceable output, reducing the time for manual chapter reading and standard search, shortening the review cycle and improving review efficiency. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating an optional intelligent review agent construction method for flood impact assessment reports provided in this application embodiment; Figure 2 A flowchart illustrating another optional method for constructing an intelligent review agent for flood impact assessment reports, provided in this application embodiment; Figure 3 A schematic diagram illustrating an optional multi-source document structuring and review object construction process provided in this application embodiment; Figure 4 A schematic diagram of an optional adaptive segmentation, evidence retrieval, and review task routing process provided for embodiments of this application; Figure 5 A schematic diagram of an optional multi-model, multi-view collaborative review and dynamic fusion process provided for an embodiment of this application; Figure 6 A schematic diagram of an optional cross-task consistency verification and traceability result generation process provided for an embodiment of this application; Figure 7 This is a schematic diagram of an optional electronic device provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0022] The terms "first," "second," "third," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0023] According to one aspect of the embodiments of this application, a method for constructing an intelligent review agent for flood impact assessment reports is provided, applied to a three-level hierarchical architecture including a first-level edge cloud management cluster, a second-level terminal cloud management cluster, and a third-level terminal node. The following describes the method in conjunction with... Figure 1 This application describes a method for constructing an intelligent auditing agent for flood impact assessment reports, as provided in its embodiments.

[0024] Figure 1 This is a flowchart illustrating an optional intelligent review agent construction method for flood impact assessment reports provided in an embodiment of this application. Figure 1 As shown, the process of this method may include the following steps: S102, Obtain input materials, including the flood impact assessment report to be reviewed, technical standard documents, regulatory guidelines documents, and supplementary review requirements; S104, The input material is structured to generate a unified review knowledge unit; S106, based on the candidate block boundaries, the unified review knowledge unit is adaptively divided into multiple review blocks, and the corresponding original report evidence and standard provisions are retrieved for each review block and assembled into an evidence context. S108, establish a capability profile of the callable model, calculate the task suitability based on the capability requirement weight of the review task and the model capability profile, dynamically select a set of models whose suitability meets the preset conditions, and under the constraints of the evidence context, conduct independent review of each review block from multiple review perspectives to obtain the review opinions of each model, perform dynamic weight fusion and conflict identification on the review opinions of the multiple models, and generate a fused review opinion. S110 performs cross-task consistency verification on the integrated review opinions of multiple review blocks, including cross-chapter parameter consistency verification, issue merging and risk ranking, and generates the final review result.

[0025] This invention provides a method for constructing an intelligent review agent for flood impact assessment reports, which is applicable to scenarios involving automated intelligent review of report documents, such as administrative approval scenarios for flood impact assessment reports by water conservancy administrative departments, report review scenarios by technical review units, etc., and can also be used for other scenarios, which are not limited here.

[0026] Combination Figure 1 , Figure 2 and Figure 3 As shown, the flood impact assessment reports to be reviewed in the input materials are flood impact assessment reports for river-related construction projects that require review. Their content typically includes a project overview, basic information on rivers, lakes, and flood storage areas, flood control standards, hydrological analysis, backwater impact, scour impact, impact on dike and bank stability, impact assessment, measures to eliminate and mitigate impacts, conclusions, and recommendations. Technical standard documents include national and industry standards related to flood impact assessment. Regulatory guidance documents include regulations, guidelines, and review points issued by the Ministry of Water Resources and relevant competent authorities. Supplementary review requirements are additional review requirements or key concerns specified by the user based on the specific project circumstances.

[0027] Furthermore, the flood impact assessment report to be reviewed undergoes text extraction, layout analysis, page number positioning, title hierarchy identification, table extraction, and special symbol cleaning to form the basic text units of the report.

[0028] Based on title numbering, paragraph style, keywords, and semantic continuity, identify the report's chapter paths, including project overview, basic information on rivers, lakes, and flood storage areas, flood control standards, hydrological analysis, impact of water backlog, scouring impact, impact on dike and bank stability, impact assessment, measures to eliminate and mitigate impacts, conclusions, and recommendations, and construct the report's chapter tree.

[0029] The technical standards, regulations, and guidelines are processed by identifying clause numbers, extracting clause content, marking applicable objects, marking the degree of mandatory nature, generating keywords, and constructing clause citation relationships to form standard clause knowledge units.

[0030] Finally, the basic text units of the report, standard provisions, and review requirements are mapped to unified review knowledge units. Each unified review knowledge unit includes a unit number, text content, content type, source location, chapter path, association rules, and semantic representation. The semantic representation characterizes the semantic features of this review knowledge unit in the review task, enabling relevance matching, task routing, and issue merging of report content, standard provisions, and review requirements within a unified semantic space.

[0031] In an exemplary embodiment, the unified review knowledge unit is represented as:

[0032] in, Indicates the first i A unified knowledge unit for review Indicates a unique unit number. Represents text content, Indicates the content type. Indicates the source location. Indicates the chapter path, Indicates association rules, Represents a semantic vector.

[0033] like Figure 4 As shown, in an exemplary embodiment, the adaptive segmentation of the unified review knowledge unit based on candidate segmentation boundaries includes: Using chapter titles, paragraph boundaries, table boundaries, calculation explanation paragraphs, evaluation conclusion paragraphs, and measure description paragraphs as candidate block boundaries, the comprehensive cost of each candidate block is calculated, expressed as: Cost(C' k )=αL(C' k )+βS(C' k )+γH(C' k ); Among them, C' k This represents the k-th candidate block, Cost(C' k L(C') represents the block total cost of the k-th candidate block. k S(C') represents the cost of exceeding the length limit. k H(C') represents the semantic fragmentation cost. k ) represents the cost of chapter-level segmentation, and α, β, and γ are weighting coefficients; The optimal block partitioning scheme is determined from the candidate blocks with the goal of minimizing the overall cost of block partitioning.

[0034] Optimal partitioning scheme It can be represented as:

[0035] in, These are candidate block partitioning schemes.

[0036] Optionally, multiple review blocks are obtained based on the optimal block partitioning scheme. The final review block is a collection of multiple consecutive review knowledge units, which satisfies the context length constraint while maintaining chapter and semantic continuity.

[0037] Based on the chapter structure, project type, and technical standard clauses of the flood impact assessment report, the review items include the completeness of the project overview, the rationality of the flood control standards, the sufficiency of the hydrological analysis, the impact analysis on the stability of the dikes and banks, the impact analysis on backwater and scour, the feasibility of remedial measures, the consistency of conclusions, the accuracy of the cited basis, and the standardization of the format.

[0038] Perform review task routing. Based on the content type and review items, tasks are routed to the corresponding agents, such as technical review, standards compliance review, formal logic review, or data consistency review.

[0039] In one exemplary embodiment, the step of retrieving the corresponding original report evidence and standard provisions for each review block includes: Combine the review blocks of text and review items into review tasks; Candidate evidence units related to the review task are retrieved from the standard clause knowledge unit set and the review knowledge unit set to form a candidate evidence set; The retrieval score of each candidate evidence unit in the candidate evidence set is calculated based on the weighted sum of semantic relevance score, keyword matching score and metadata matching score, and candidate evidence units whose retrieval scores meet the preset conditions are selected as the basis for review. The metadata includes at least one of the following: chapter path, content type, page number position, review topic tags, scope of application, or clause type.

[0040] In one specific implementation, the evidence retrieval score is represented as:

[0041]

[0042]

[0043] Among them, the review task By the Each review block and the One review item Together constitute Indicates a combination of query information; This represents candidate evidence units, which are derived from the standard clause knowledge unit set. Or examine the set of knowledge units R (i.e. (a set); This represents the semantic relevance score between the retrieval query and the candidate evidence unit; Indicates the keyword matching score; This indicates the metadata matching score, whereby the metadata includes at least one of the following: chapter path, content type, page number position, review topic tags, scope of application, or clause type. , , These are the weighting coefficients.

[0044] The selected candidate evidence units are assembled into evidence context for use in subsequent review steps.

[0045] Optionally, such as Figure 5 As shown, the callable model records indicators such as professional technical understanding ability, standard clause understanding ability, long text processing ability, logical consistency judgment ability, historical review accuracy rate, and call response latency.

[0046] For each review task, a capability requirement weight vector is generated based on the capability dimensions required for the review items, and the task fit of candidate models is calculated in combination with the model capability profile; then, a set of models with high task fit is selected from the callable models to participate in multi-perspective review.

[0047] In one exemplary embodiment, the task adaptability is represented as:

[0048] in, Indicates a callable model. This represents the k-th review block. This indicates the d-th review item. Represents a set of capabilities. This represents the weight of the current task's requirement for the r-th type of capability. Indicates a callable model The score on the r-th ability Indicates a callable model Response latency, These are the weighting coefficients.

[0049]

[0050] Represents the candidate pool of the model. Indicates that for the first k The first block and the first d The set of models selected from each review item.

[0051] Each model outputs review opinions for each review task from the perspectives of professional and technical review, standard compliance review, formal logic review, data consistency review, and risk classification review, under the same evidence context.

[0052] Each review comment should be checked to ensure it includes the original report evidence, the relevant standard provisions, the rationale for the judgment, and suggested modifications. Result lacking supporting evidence will be downgraded or trigger a re-review.

[0053] In an exemplary embodiment, the dynamic weight fusion of the review opinions of the plurality of models includes: The review opinions generated by each model are normalized, and the conclusion judgments are mapped to the candidate conclusion set; The fusion score of each candidate conclusion is calculated by combining the historical capability score of the integrated model, the evidence coverage of the current output, the consistency score with the conclusions of other models, and the model's self-assessed confidence. The review opinion corresponding to the candidate conclusion with the highest fusion score is used as the fusion review opinion.

[0054] Specifically, the model fusion weights can be expressed as:

[0055] in, Indicates a callable model m The fusion weight, This represents the mean of the weights across all perspectives. Indicates a callable model m From the current perspective q Historical accuracy Indicates a callable model m in Current perspective q The output evidence coverage, Indicates a callable model m Consistency score with conclusions from other models This represents all candidate models (including callable models and other models). Represents all candidate models From the current perspective q Historical accuracy Represents all candidate models From the current perspective q The output evidence coverage, Represents all candidate models Consistency score with conclusions from other models , , This is the adjustment coefficient.

[0056] The fusion score of the candidate conclusions is expressed as follows:

[0057] in, The fusion score represents the candidate conclusion y. As an indicator function, when the callable model m Conclusion The value is 1 if it equals the candidate conclusion y, and 0 otherwise. This represents the model's self-assessed confidence level. The final fusion conclusion is the candidate conclusion with the highest score.

[0058] In one exemplary embodiment, the method further includes: During the dynamic weighted fusion of the review opinions from the multiple models, the degree of conflict between the conclusions of the review opinions generated by different models is calculated. When the degree of conflict of the conclusion exceeds the preset threshold, the basis is retrieved again and a second review is conducted, or the review task is marked as pending manual review.

[0059] After conflict identification and secondary adjudication, a fusion review opinion is generated. Optionally, based on the fusion conclusion, the following can be output: issue name, issue location, issue description, original report reference, standard clause reference, risk level, and modification suggestions.

[0060] Furthermore, such as Figure 6 As shown, the task review results are summarized. All task outputs, including questions, supporting evidence, risk levels, and recommendations, are compiled into a candidate question set.

[0061] Merge and deduplicate questions. Based on semantic similarity of question descriptions, similarity of citations, and similarity of chapter positions, merge questions that are identical or highly similar.

[0062] Question similarity can be represented as:

[0063] in, Indicate the problem and The overall similarity Indicates the semantic similarity of the problem descriptions. This indicates that the citation is based on similarity. Indicates the similarity of chapter positions. , , These are the weighting coefficients.

[0064] Extract key entities and parameters such as project name, construction location, river name, design flood standard, design flow, design water level, coordinates, elevation, project dimensions, remedial measures, and evaluation conclusions.

[0065] Perform cross-chapter consistency checks. Compare the values, units, expressions, and sources of the same parameter in different chapters, tables, and conclusions to identify inconsistencies or insufficient evidence for conclusions.

[0066] Risk scoring and ranking are conducted. Risk scores are calculated based on indicators such as the severity of the problem, its impact on flood control safety and approval compliance, and the degree of missing data, and are categorized into major problems, general problems, suggestive problems, and format-related problems.

[0067] The risk score can be expressed as:

[0068] in, Indicate the problem Risk score, Indicates the severity. This indicates the degree of impact on flood control safety and approval compliance. Indicates the degree of missing data or insufficient evidence. , , These are the weighting coefficients.

[0069] Generate the final review results. The final results include the overall review conclusions, a list of issues, the relevant chapters of the issues, original evidence from the report, the basis for the standard provisions, the risk level, suggested revisions, and the review process log.

[0070] Receive feedback from human reviewers. Human reviewers confirm, modify, reject, or supplement the issues, and the system updates the review rules, model capability profile, and dynamic fusion weights based on the feedback.

[0071] The updated model capability score after human feedback can be represented as:

[0072] in, This represents the updated model capability score. This represents the model's capability score before the update. This represents the performance score obtained from human feedback. This indicates the update step size.

[0073] Through steps S102 to S110, input materials are acquired, structured, and a unified review knowledge unit is generated. Based on candidate block boundaries, the unified review knowledge unit is adaptively divided into multiple review blocks. For each review block, corresponding original report evidence and standard provisions are retrieved and assembled into an evidence context. A capability profile of the callable model is established. The task suitability is calculated based on the capability requirement weights of the review task and the model capability profile. A set of models whose suitability meets preset conditions is dynamically selected. Under the constraints of the evidence context, each review block is reviewed from multiple review perspectives, obtaining review opinions from each model. Dynamic weight fusion and conflict identification are performed on the review opinions from multiple models to generate a fused review opinion. Cross-task consistency verification is performed on the fused review opinions from multiple review blocks to generate the final review result. This solves the technical problems in existing flood impact assessment report review, such as low review efficiency, complex standard reference retrieval, difficulty in discovering cross-chapter parameter contradictions, and lack of stability and interpretability of review results due to reliance on manual labor or a single model.

[0074] According to another aspect of the embodiments of this application, a system for constructing an intelligent agent for reviewing flood impact assessment reports is also provided, which executes the above-described method for constructing an intelligent agent for reviewing flood impact assessment reports, including: The acquisition module is used to acquire input materials, including flood impact assessment reports to be reviewed, technical standard documents, regulatory guidelines, and supplementary review requirements. The structuring module is used to perform structuring processing on the input materials to generate unified review knowledge units; The segmentation and retrieval module is used to adaptively segment the unified review knowledge unit based on the candidate segmentation boundary to obtain multiple review segments. For each review segment, the module retrieves the corresponding original report evidence and standard provisions, and assembles them into an evidence context. The multi-model collaborative review module is used to establish a capability profile of callable models, calculate the task suitability based on the capability requirement weight of the review task and the model capability profile, dynamically select a set of models whose suitability meets preset conditions, and independently review each review block from multiple review perspectives under the constraints of the evidence context to obtain the review opinions of each model. The review opinions of the multiple models are dynamically weighted and conflict identified to generate a fused review opinion. The cross-task verification module is used to perform cross-task consistency verification on the integrated review opinions of multiple review blocks, including cross-chapter parameter consistency verification, issue merging and risk ranking, and generate the final review result.

[0075] For example, users upload flood impact assessment reports, required technical standards and regulatory guidelines, and supplementary review points, and set the block length. The system first parses the report's chapter tree and standard clause library, dividing the report content into several review blocks; then, it retrieves relevant standard bases based on the chapter and review list of each block; next, multiple models with different capabilities conduct professional technical reviews, standard compliance reviews, formal logic reviews, data consistency reviews, and risk classification reviews; the system integrates the adjudication process to perform evidence constraint verification and dynamic weighted fusion of the multi-model outputs; finally, it performs cross-chapter parameter consistency verification and problem merging on all block results, and outputs the final review report.

[0076] In the above implementation examples, the multi-model collaborative review process does not limit the specific model names or the number of participating models. The system can dynamically adjust the model combination based on model availability, review task complexity, cost constraints, and review risk level. For high-risk or highly conflict-ridden review items, a second round of adjudication or manual review can be added; for low-risk format-related issues, lightweight models or rule-based validation methods can be used.

[0077] This embodiment ensures that the segmentation satisfies the model context constraints while maintaining the integrity and semantic continuity of the chapters as much as possible. This reduces the risk of unstable review results or insufficient professional judgment from a single model, and improves the accuracy and reliability of the review conclusions.

[0078] According to another aspect of the embodiments of this application, a storage medium is also provided. Optionally, in this embodiment, the storage medium can be used to execute the program code of the intelligent agent construction method for flood impact assessment report intelligent review described in any of the embodiments of this application.

[0079] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: S1. Obtain input materials, including the flood impact assessment report to be reviewed, technical standard documents, regulatory guidelines documents, and supplementary review requirements; S2, perform structured processing on the input material to generate unified review knowledge units; S3, based on the candidate block boundaries, the unified review knowledge unit is adaptively divided into multiple review blocks, and the corresponding original report evidence and standard provisions are retrieved for each review block and assembled into an evidence context. S4. Establish a capability profile of the callable model, calculate the task suitability based on the capability requirement weight of the review task and the model capability profile, dynamically select a set of models whose suitability meets the preset conditions, and independently review each review block from multiple review perspectives under the constraints of the evidence context to obtain the review opinions of each model. Perform dynamic weight fusion and conflict identification on the review opinions of the multiple models to generate a fused review opinion. S5 performs cross-task consistency checks on the merged review opinions from multiple review blocks, including cross-chapter parameter consistency checks, issue merging, and risk ranking, and generates the final review results.

[0080] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated in this embodiment.

[0081] The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0082] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described intelligent review agent construction method for flood impact assessment reports is also provided. The electronic device may be a server, a terminal, or a combination thereof.

[0083] Figure 7 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application, such as... Figure 7 As shown, it includes a processor 702, a communication interface 704, a memory 706, and a communication bus 708. The processor 702, communication interface 704, and memory 706 communicate with each other via the communication bus 708. Memory 706 is used to store computer programs; When processor 702 executes a computer program stored in memory 706, it performs the following steps: S1. Obtain input materials, including the flood impact assessment report to be reviewed, technical standard documents, regulatory guidelines documents, and supplementary review requirements; S2, perform structured processing on the input material to generate unified review knowledge units; S3, based on the candidate block boundaries, the unified review knowledge unit is adaptively divided into multiple review blocks, and the corresponding original report evidence and standard provisions are retrieved for each review block and assembled into an evidence context. S4. Establish a capability profile of the callable model, calculate the task suitability based on the capability requirement weight of the review task and the model capability profile, dynamically select a set of models whose suitability meets the preset conditions, and independently review each review block from multiple review perspectives under the constraints of the evidence context to obtain the review opinions of each model. Perform dynamic weight fusion and conflict identification on the review opinions of the multiple models to generate a fused review opinion. S5 performs cross-task consistency checks on the merged review opinions from multiple review blocks, including cross-chapter parameter consistency checks, issue merging, and risk ranking, and generates the final review results.

[0084] Optionally, the communication bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 The symbol is represented by a single thick line, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned electronic device and other devices.

[0085] The memory may include RAM, or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0086] The processors mentioned above can be general-purpose processors, including but not limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; they can also be DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0087] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0088] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0089] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0090] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0091] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

[0092] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing an intelligent review agent for flood impact assessment reports, characterized in that, include: Obtain input materials, including the flood impact assessment report to be reviewed, technical standard documents, regulatory guidelines, and supplementary review requirements; The input materials are structured to generate unified review knowledge units; Based on the candidate block boundaries, the unified review knowledge unit is adaptively divided into multiple review blocks. For each review block, the corresponding original report evidence and standard provisions are retrieved and assembled into an evidence context. Establish a capability profile of the callable model, calculate the task suitability based on the capability requirement weight of the review task and the model capability profile, dynamically select a set of models whose suitability meets the preset conditions, and independently review each review block from multiple review perspectives under the constraints of the evidence context to obtain the review opinions of each model. Perform dynamic weight fusion and conflict identification on the review opinions of the multiple models to generate a fused review opinion. Perform cross-task consistency verification on the integrated review opinions from multiple review blocks, including cross-chapter parameter consistency verification, issue merging and risk ranking, and generate the final review result.

2. The method for constructing an intelligent review agent for flood impact assessment reports as described in claim 1, characterized in that, The unified review knowledge unit is represented as follows: in, Indicates the first i A unified knowledge unit for review Indicates a unique unit number. Represents text content, Indicates the content type. Indicates the source location. Indicates the chapter path, Indicates association rules, Represents a semantic vector.

3. The method for constructing an intelligent review agent for flood impact assessment reports as described in claim 1, characterized in that, The adaptive segmentation of the unified review knowledge unit based on candidate segmentation boundaries includes: Using chapter titles, paragraph boundaries, table boundaries, calculation explanation paragraphs, evaluation conclusion paragraphs, and measure description paragraphs as candidate block boundaries, the comprehensive cost of each candidate block is calculated, expressed as: Cost(C' k )=αL(C' k )+βS(C' k )+γH(C' k ); Among them, C' k This represents the k-th candidate block, Cost(C' k L(C') represents the block total cost of the k-th candidate block. k S(C') represents the cost of exceeding the length limit. k H(C') represents the semantic fragmentation cost. k ) represents the cost of chapter-level segmentation, and α, β, and γ are weighting coefficients; The optimal block partitioning scheme is determined from the candidate blocks with the goal of minimizing the overall cost of block partitioning.

4. The method for constructing an intelligent review agent for flood impact assessment reports as described in claim 1, characterized in that, The process involves retrieving the original report evidence and standard provisions corresponding to each review segment, including: Combine the review blocks of text and review items into review tasks; Candidate evidence units related to the review task are retrieved from the standard clause knowledge unit set and the review knowledge unit set to form a candidate evidence set; The retrieval score of each candidate evidence unit in the candidate evidence set is calculated based on the weighted sum of semantic relevance score, keyword matching score and metadata matching score, and candidate evidence units whose retrieval scores meet the preset conditions are selected as the basis for review. The metadata includes at least one of the following: chapter path, content type, page number position, review topic tags, scope of application, or clause type.

5. The method for constructing an intelligent review agent for flood impact assessment reports as described in claim 1, characterized in that, The dynamic weight fusion of review opinions from the multiple models includes: The review opinions generated by each model are normalized, and the conclusion judgments are mapped to the candidate conclusion set; The fusion score of each candidate conclusion is calculated by combining the historical capability score of the integrated model, the evidence coverage of the current output, the consistency score with the conclusions of other models, and the model's self-assessed confidence. The review opinion corresponding to the candidate conclusion with the highest fusion score is used as the fusion review opinion.

6. The method for constructing an intelligent review agent for flood impact assessment reports as described in claim 5, characterized in that, The method further includes: During the dynamic weighted fusion of the review opinions from the multiple models, the degree of conflict between the conclusions of the review opinions generated by different models is calculated. When the degree of conflict of the conclusion exceeds the preset threshold, the basis is retrieved again and a second review is conducted, or the review task is marked as pending manual review.

7. The method for constructing an intelligent review agent for flood impact assessment reports as described in claim 1, characterized in that, The task suitability is expressed as: in, Indicates a callable model. This represents the k-th review block. This indicates the d-th review item. Represents a set of capabilities. This represents the weight of the current task's requirement for the r-th type of capability. Indicates a callable model The score on the r-th ability Indicates a callable model Response latency, These are the weighting coefficients.

8. A system for constructing an intelligent review agent for flood impact assessment reports, characterized in that, include: The acquisition module is used to acquire input materials, including flood impact assessment reports to be reviewed, technical standard documents, regulatory guidelines, and supplementary review requirements. The structuring module is used to perform structuring processing on the input materials to generate unified review knowledge units; The segmentation and retrieval module is used to adaptively segment the unified review knowledge unit based on the candidate segmentation boundary to obtain multiple review segments. For each review segment, the module retrieves the corresponding original report evidence and standard provisions, and assembles them into an evidence context. The multi-model collaborative review module is used to establish a capability profile of callable models, calculate the task suitability based on the capability requirement weight of the review task and the model capability profile, dynamically select a set of models whose suitability meets preset conditions, and independently review each review block from multiple review perspectives under the constraints of the evidence context to obtain the review opinions of each model. The review opinions of the multiple models are dynamically weighted and conflict identified to generate a fused review opinion. The cross-task verification module is used to perform cross-task consistency verification on the integrated review opinions of multiple review blocks, including cross-chapter parameter consistency verification, issue merging and risk ranking, and generate the final review result.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method of any one of claims 1 to 7 through the computer program.

Images