A method and device for bridge tunnel group maintenance priority sequencing and explanation generation

Abstract

The application discloses a bridge and tunnel group maintenance priority ranking and explanation generation method and device, and relates to the technical field of civil engineering and artificial intelligence. The method comprises the following steps: generating a bridge and tunnel annual state image; processing unstructured external basis and expert guidance related data to form knowledge units and expert guidance units, and constructing a bridge and tunnel maintenance knowledge base; inputting the state image, the expert guidance unit, the evidence package and the version identifier into a fixed template to organize a search enhanced large model, and outputting a Rubric evaluation criterion, a candidate ranking explanation, cited evidence and artificial review marks; and calculating a score to generate a bridge and tunnel maintenance priority ranking table. The application can make the bridge and tunnel group maintenance priority ranking result simultaneously have structured data support, expert rule support, knowledge evidence support and report explanation support, and provide auxiliary basis for pre-ranking review, management report and artificial decision of the annual maintenance plan.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of civil engineering and artificial intelligence, and in particular to a method and apparatus for generating and interpreting the maintenance priorities of bridge and tunnel groups. Background Technology

[0002] Bridges, tunnels, and bridge-tunnel complexes are critical infrastructure in urban road traffic systems and regional transportation networks. Their structural safety, traffic capacity, and service levels directly affect urban traffic operation, public safety, emergency response, and regional economic activities. With the increasing number of urban bridge and tunnel assets, their service life, changes in traffic loads, and the growing impact of extreme weather and emergencies, bridge and tunnel maintenance management has gradually shifted from the inspection, evaluation, and repair of individual structures to annual condition assessments, risk identification, prioritization, and management reporting for bridge and tunnel complexes. For management departments, annual maintenance work is usually not carried out in isolation on a single bridge or tunnel, but requires a comprehensive comparison of a large number of bridge and tunnel assets within a given year to identify those requiring focused attention, priority review, or inclusion in subsequent maintenance plans. Therefore, how to formulate an interpretable, traceable, and verifiable annual maintenance priority ranking based on bridge and tunnel technical condition, defect information, operational functions, road network location, historical maintenance records, and expert opinions has become a significant technical issue in bridge and tunnel complex maintenance management. Current bridge and tunnel maintenance management is primarily based on industry norms and technical standards. The "Specifications for Highway Bridge and Culvert Maintenance" (JTG 5120—2021) sets forth requirements for the inspection, evaluation, repair, and technical management of highway bridges and culverts, providing an industry basis for bridge and culvert maintenance management. The "Technical Specifications for Highway Tunnel Maintenance" (JTG H12—2015) stipulates the maintenance levels, inspection, evaluation, repair procedures, and technical management of highway tunnels, providing a basic standard for tunnel maintenance management. The "Technical Standards for Urban Bridge Maintenance" (CJJ 99—2017) applies to the maintenance of urban bridges delivered for use after completion and acceptance, emphasizing the maintenance of urban bridge facilities and ensuring the integrity and safe operation of bridges. These standards and specifications provide a rule-based foundation for bridge and tunnel inspection, evaluation, maintenance management, and technical procedures. However, the specifications themselves primarily answer the questions of "how to inspect, how to evaluate, and how to manage," without directly providing a method for multi-source information fusion, expert opinion absorption, priority ranking, and report generation for annual bridge and tunnel groups. From the perspective of bridge management system development, bridge maintenance management has gradually evolved from early asset ledgers, inspection records, and single-bridge evaluations to a bridge management system and maintenance auxiliary decision-making system oriented towards bridge asset groups. The Pontis system was among the first to incorporate bridge network maintenance optimization and funding allocation into a systematic management framework, laying a crucial foundation for the development of bridge management systems. Bridge management systems further integrate the condition assessment, maintenance needs, and maintenance optimization of existing bridges, reflecting a trend of expansion from data management to maintenance decision support. A review of the current state of bridge management systems in the digital age indicates that they typically involve modules such as data management, diagnosis, prediction, and decision-making, reflecting the ongoing development of bridge management towards digitalization, informatization, and intelligence.These studies demonstrate that bridge management systems can provide a foundation for data organization, condition assessment, and decision support for bridge and tunnel asset maintenance. However, in the annual bridge and tunnel group ranking scenario, further solutions are needed for issues such as how to uniformly represent multi-source data, how to structure expert opinions for inclusion in the ranking process, how to automatically interpret the ranking criteria, and how to trace and verify the interpretation content. Bridge priority ranking should not be based solely on a single technical condition level but should comprehensively consider structural condition, risk consequences, road network function, and management objectives. However, the interpretation of bridge priority ranking typically relies on indicator systems, indicator weights, and comprehensive scores, with little further explanation of the original data basis, expert opinion basis, standard knowledge basis, and sources of uncertainty corresponding to each ranking result. In engineering practice, annual bridge and tunnel maintenance ranking not only relies on structured indicators but also heavily depends on expert experience. Judgments based on expert experience often originate from expert meetings, review opinions, maintenance experience, management requirements, or annual key tasks, possessing significant engineering value. However, these judgments are usually presented in natural language, making them difficult to calculate directly using traditional scoring models and difficult to reliably trace in ranking reports. Summary of the Invention

[0003] To address the problems in existing retrieval-enhanced large language models, such as the tendency for illusions or inconsistent outputs during the retrieval and generation phases (including incomplete retrieval results, contextual noise, contextual conflicts, evidence mismatch, and model-generated content deviating from input evidence), and the difficulty in integrating multi-source data, structuring expert opinions, and tracing the ranking criteria in bridge and tunnel group maintenance prioritization, this invention provides a method and apparatus for generating bridge and tunnel group maintenance priority ranking and interpretation. The technical solution is as follows:

[0004] On the one hand, a method for prioritizing and generating interpretations of bridge and tunnel group maintenance is provided. This method is implemented by a bridge and tunnel group maintenance priority prioritization and interpretation generation device, and includes: S1: Collect annual status data of bridges and tunnels, perform data aggregation, standardization processing, record merging, missing field marking and profile generation processing to obtain the annual status profile module corresponding to each bridge or tunnel, and generate an annual status profile based on the annual status profile module; the annual status profile includes asset identifier, annual identifier, status field, tag field, missing field, source reference and confidence mark; S2: Collect external evaluation materials and unstructured data sources based on expert guidance, establish source records and segment semantic boundaries to obtain external basis knowledge units; at the same time, convert the unstructured data sources based on expert guidance into structured conditions that can match the fields of the annual status profile to obtain expert guidance units. S3: Extract the evaluation basis version identifier through external basis knowledge units and expert guidance units; construct a knowledge base based on the annual status profile, external basis knowledge units and expert guidance units to obtain the bridge and tunnel maintenance knowledge base and the new evaluation basis version identifier; S4: Based on the applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit, retrieve candidate evidence from the bridge and tunnel maintenance knowledge base to obtain a set of candidate evidence; based on the screening rules, screen the candidate evidence set according to asset applicability, source reliability, and content relevance to obtain a bridge and tunnel level knowledge evidence package, and mark it as missing evidence. S5: Using a fixed prompt template, organize the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier, and inference constraints to obtain the input content of the retrieval enhancement model; input the input content of the retrieval enhancement model into the retrieval enhancement model to obtain the output content, which includes the asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking interpretation results, cited evidence number, and manual review mark output; S6: Using the output content of the retrieval enhancement model, external basis knowledge units, and expert guidance units, the consistency of the output of the retrieval enhancement model is verified to obtain the Rubric evaluation criteria, Rubric dimension weights, and verification marks that pass the verification; finally, a comprehensive score is calculated to form a bridge and tunnel maintenance priority ranking table.

[0005] Preferably, the annual status data of bridges and tunnels collected in S1 is processed through data aggregation, standardization, record merging, missing field marking, and profile generation to obtain an annual status profile module corresponding to each bridge or tunnel. An annual status profile is then generated based on this module. The annual status profile includes asset identifiers, annual identifiers, status fields, tag fields, missing fields, source references, and confidence level markers, including: S11: Collect annual status data of bridges and tunnels. The annual status data of bridges and tunnels includes bridge and tunnel asset ledgers, inspection and evaluation data, defect records, maintenance records, traffic and road network information and manual supplementary records. The bridge and tunnel asset ledgers include, but are not limited to, the main index and asset names. S12: Using the asset number as the primary index, collect the annual status data of the bridge and tunnel; S13: When asset numbers are missing or inconsistent, the asset name, the road where it is located, and the spatial location auxiliary fields are combined for matching to obtain asset alignment records; S14: Standardize the asset alignment records, merge the records, and mark missing fields to obtain structured bridge and tunnel annual status data; the standardization process includes unifying units and checking for outliers in numerical fields and extracting keywords and tags from text fields; the record merging process includes merging according to time, severity, and data source reliability; and the missing field marking process includes explicitly marking the missing key fields. S15: Based on the structured annual status data of bridges and tunnels, the tasks are sorted according to the priority of bridge and tunnel maintenance and processed through portrait generation to obtain the annual status portrait module corresponding to each bridge or tunnel, and the original source corresponding to the key fields is retained; the annual status portrait module includes basic attribute unit, technical condition unit, defect feature unit, maintenance history unit, traffic function unit, management concern unit and data integrity unit. The annual status portrait module is used to generate the annual status portrait, which includes asset identifier, annual identifier, status field, tag field, missing field, source reference and confidence mark.

[0006] Preferably, in step S2, the external evaluation materials and unstructured data sources based on expert guidance are collected, and source records are established and semantic boundaries are segmented to obtain external basis knowledge units. Simultaneously, the unstructured data sources based on expert guidance are converted into structured conditions that can match the fields of the annual status profile to obtain expert guidance units, including: S21: Collect external evaluation materials and unstructured data sources based on expert guidance; S22: Using external evaluation materials, source records are established and semantic boundaries are segmented to obtain external basis knowledge units; the external evaluation materials include bridge and tunnel inspection and evaluation standards, local management documents and annual management requirements, and the external basis knowledge units record knowledge number, source material, applicable object, triggering condition, knowledge content and confirmation status. S23: Based on the unstructured data source based on expert guidance, extract structured conditions that can match the annual status profile field to obtain the expert guidance unit; the expert guidance unit is used to express the manual attention rules and review requirements in the target annual maintenance priority ranking.

[0007] Preferably, in step S3, the evaluation basis version identifier is extracted through external knowledge units and expert guidance units; a knowledge base is constructed based on the annual status profile, external knowledge units, and expert guidance units to obtain a bridge and tunnel maintenance knowledge base and a new evaluation basis version identifier, including: S31: By using external knowledge units and expert guidance units, extract the material number, knowledge unit number, source information and update time to form a version text, and perform hash calculation or equivalent unique identifier calculation on the version text to obtain the evaluation basis version identifier; S32: Based on the annual status profile, structured external basis knowledge units, and expert guidance units, a knowledge base is constructed to obtain the bridge and tunnel maintenance knowledge base; S33: Based on the evaluation basis version identifier, a version change judgment is made to obtain version constraints for subsequent retrieval enhancement big model reasoning; the version change judgment includes: when the external evaluation basis version identifier has not changed, the retrieval enhancement big model reasoning is continued under the same basis version constraint; when the external evaluation basis materials, unstructured data sources based on expert guidance opinions, or annual management priorities change, a new evaluation basis version identifier is regenerated, and the new evaluation basis version identifier is used to re-execute bridge and tunnel level knowledge evidence package retrieval and retrieval enhancement big model reasoning.

[0008] Preferably, in step S4, based on the applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit, a search is conducted from the bridge and tunnel maintenance knowledge base to obtain a set of candidate evidence; based on screening rules, the candidate evidence set is screened according to asset applicability, source reliability, and content relevance to obtain a bridge and tunnel-level knowledge evidence package, and missing evidence is marked, including: S41: Based on the applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit, a combination of field condition matching, keyword matching, and semantic retrieval is used to retrieve candidate evidence from the bridge and tunnel maintenance knowledge base; the applicable conditions include asset type, technical condition, defect labels, traffic function, and management concern labels. S42: Based on the screening rules, the candidate evidence set is screened according to asset applicability, source reliability and content relevance to obtain the evidence set; the screening rules include retaining the candidate evidence with clear source, clear applicable conditions and a matching degree with the annual status profile higher than a preset threshold for content-related candidate evidence, and retaining the missing evidence mark for the case where no valid evidence is found; S43: Set an evidence number, associated source knowledge unit, and source material for each piece of evidence in the evidence set, and save the bridge and tunnel level knowledge evidence package together with the asset number and evaluation basis version identifier to obtain the bridge and tunnel level knowledge evidence package; S44: Based on the fields and source references in the annual status profile, verify the consistency between the bridge and tunnel level knowledge evidence package and the original data. The missing evidence marker is used to constrain the uncertainty prompts in the subsequent candidate ranking interpretation.

[0009] Preferably, step S5 uses a fixed prompt template to organize the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier, and inference constraints to obtain the input content of the retrieval enhancement model; the input content of the retrieval enhancement model is then input into the retrieval enhancement model to obtain the output content, which includes the asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking interpretation results, cited evidence number, and manual review mark output, including: S51: Using a fixed prompt template, the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier and reasoning constraints are organized to obtain the input content of the retrieval enhancement model. The fixed prompt template organizes the content by using a single bridge or tunnel as the processing unit. S52: Input the input content of the retrieval enhancement model into the retrieval enhancement model to obtain the output content of the retrieval enhancement model; the output content includes the first part of the content Rubric evaluation criteria and the second part of the content candidate ranking interpretation results. The first part of the content Rubric evaluation criteria is constrained by external basis knowledge units, expert guidance units, and evaluation basis version identifiers. The second part of the content candidate ranking interpretation results includes candidate scores, candidate priorities, ranking reasons, risk triggering factors, evidence number citations, and uncertainty prompts. The second part of the content candidate ranking interpretation results is obtained based on the annual status profile and bridge and tunnel level knowledge evidence package. S53: Using a fixed JSON structure, the output content of the retrieval enhancement model is saved, resulting in an output including asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking interpretation results, cited evidence number, and manual review mark.

[0010] Preferably, step S6 uses the output content of the retrieval enhancement model, external basis knowledge units, and expert guidance units to perform consistency verification on the output of the retrieval enhancement model, obtaining the verified Rubric evaluation criteria, Rubric dimension weights, and verification marks; finally, a comprehensive score is calculated to form a bridge and tunnel maintenance priority ranking table, including: S61: Using the output content of the retrieval enhancement model, external basis knowledge units, and expert guidance units, the Rubric evaluation criteria of the output content of the retrieval enhancement model are validated for consistency. The validated Rubric evaluation criteria, Rubric dimension weights, and verification marks are obtained. The consistency validation includes checking whether the evaluation dimensions, dimension weights, scoring anchors, and external basis references meet the requirements of the ranking task. When there is a missing evaluation dimension, a weight that does not meet the calculation requirements, a scoring anchor that cannot correspond to the engineering meaning, or an external basis reference that cannot be traced back to the structured knowledge unit, the corresponding Rubric content is marked as a criterion anomaly. S62: Based on the annual status profile, bridge and tunnel level knowledge evidence package and expert guidance unit, perform data and evidence verification on the candidate ranking interpretation results to obtain candidate ranking interpretation results that pass the verification; the data and evidence verification includes verifying the scoring reasons, the cited evidence number and the triggering of the expert guidance unit. When a candidate result has external input information, the evidence number does not exist or the evidence does not match the reasons, record the abnormal type and restrict it from entering the final ranking interpretation. S63: Calculate the comprehensive score based on the verified Rubric evaluation criteria, Rubric dimension weights, and review markers, and correct bridge and tunnel assets that trigger expert attention, have abnormal risks, or lack key data to obtain the corrected results; the correction includes retaining manual review markers, retaining records of score anomalies, and hiding untraceable explanatory content; S64: Based on the corrected results, prioritize bridge and tunnel maintenance to obtain a bridge and tunnel maintenance priority ranking table; the bridge and tunnel maintenance priority ranking table includes asset number, comprehensive score, priority level, ranking position and review status.

[0011] On the other hand, a bridge and tunnel group maintenance priority ranking and interpretation generation device is provided. This device is applied to the bridge and tunnel group maintenance priority ranking and interpretation generation method. The device includes: Annual Status Profile Module: This module is used to collect annual status data of bridges and tunnels, perform data aggregation, standardization, record merging, missing field marking, and profile generation to obtain the annual status profile module corresponding to each bridge or tunnel. An annual status profile is generated based on the annual status profile module. The annual status profile includes asset identifier, annual identifier, status field, tag field, missing field, source reference, and confidence level mark. External basis module: used to collect external evaluation basis materials and unstructured data sources based on expert guidance, establish source records and semantic boundary segmentation to obtain external basis knowledge units; at the same time, convert the unstructured data sources based on expert guidance into structured conditions that can match the fields of the annual status profile to obtain expert guidance units; Knowledge Base Module: Used to extract evaluation basis version identifiers through external basis knowledge units and expert guidance units; based on the annual status profile, external basis knowledge units, and expert guidance units, a knowledge base is constructed to obtain a bridge and tunnel maintenance knowledge base and new evaluation basis version identifiers; Candidate Evidence Module: This module retrieves candidate evidence from the bridge and tunnel maintenance knowledge base based on applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit. Based on screening rules, the candidate evidence set is filtered according to asset applicability, source reliability, and content relevance to obtain a bridge and tunnel level knowledge evidence package, and missing evidence is marked. Large Model Module: This module organizes the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier, and inference constraints using a fixed prompt template to obtain the input content for the retrieval enhancement large model. The input content is then input into the retrieval enhancement large model to obtain the output content, which includes the asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking explanation results, cited evidence number, and manual review mark. Priority ranking module: It is used to perform consistency verification on the output of the retrieval enhancement model by using the output content of the retrieval enhancement model, external basis knowledge units and expert guidance units, to obtain the Rubric evaluation criteria, Rubric dimension weights and verification marks that pass the verification; finally, it performs comprehensive score calculation to form a bridge and tunnel maintenance priority ranking table.

[0012] On the other hand, a bridge and tunnel group maintenance priority ranking and interpretation generation device is provided, the bridge and tunnel group maintenance priority ranking and interpretation generation device comprising: a processor; a memory, the memory storing computer-readable instructions, which, when executed by the processor, implement the method described in any one of the above-described bridge and tunnel group maintenance priority ranking and interpretation generation methods.

[0013] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: To address the issues of fragmented existing bridge and tunnel data, which require manual organization and interpretation before sorting, this invention constructs an annual status profile for bridges and tunnels. This profile organizes scattered information such as asset ledgers, inspection and evaluation data, defect records, and maintenance records into a unified, structured status record. Consequently, the structural status, defect status, maintenance history, and data completeness of a single bridge or tunnel in a target year can be expressed within a single object, reducing the reliance on manual searching, comparison, and summarization before sorting.

[0014] To address the problem that traditional sorting methods struggle to incorporate expert opinions in natural language and external management requirements, this invention structures expert guidance, standards, regulations, and other materials into expert guidance units and external basis knowledge units. This structured result retains records of applicable objects, triggering conditions, and sources, enabling expert opinions and external evaluation bases to participate in sorting reasoning, evidence retrieval, and report traceability processes, rather than merely existing as manual notes or post-hoc explanations.

[0015] To address the issues of unclear evidence, ambiguous citations, and extraneous information when directly generating ranking explanations using large language models, this invention constructs a bridge and tunnel maintenance knowledge base before the large model's inference process, and retrieves and forms a bridge-tunnel level knowledge evidence package for each bridge or tunnel. This evidence package, as part of the large model's input, provides a numbered range of evidence for candidate ranking reasons, enabling the model-generated explanations to cite clear knowledge sources and facilitating subsequent verification.

[0016] To address the shortcomings of traditional multi-indicator ranking methods, such as insufficient explanation of results and a tendency for evaluation criteria to deviate from the rationale behind the ranking, this invention simultaneously generates Rubric evaluation criteria and candidate ranking explanations during the retrieval enhancement model inference process. Evaluation dimensions, dimension weights, scoring anchors, candidate scores, and ranking rationales are formed within the same external basis version and evidentiary context, enabling the ranking process to not only provide scores and grades but also explain the corresponding evaluation scale and source of evidence.

[0017] To address the lack of engineering verification mechanisms in existing large language model report generation, this invention establishes a consistency verification and priority correction process for ranking results, performing procedural checks on the evaluation criteria, candidate scores, ranking reasons, and evidence citations output by the large model. For facts not present in the state profile, evidence numbers not present in the evidence package, and candidate results inconsistent with expert guidance units, the system can record anomalies, restrict their entry into the final interpretation, and retain a manual review mark.

[0018] To address the issue that existing sorting systems typically only output sorting tables, making it difficult to support management review and reporting, this invention generates a maintenance priority sorting explanation report. This report retains the maintenance priority sorting table, overall score, priority level, main sorting basis, evidence citations, uncertainty alerts, and manual review items, enabling management to review the sorting basis, identify key bridge and tunnel assets, and pinpoint areas requiring supplementary verification before the annual maintenance plan is finalized.

[0019] To address the issue of confusion between maintenance prioritization and the final maintenance decision boundary, this invention limits the output to a maintenance priority ranking and explanation report, without directly generating specific maintenance actions, methods, or budget-constrained maintenance project combinations. This boundary makes the invention applicable to the prioritization, review, and management reporting stages before the annual bridge and tunnel maintenance plan is prepared, while preserving the engineer's judgment space for the final maintenance decision. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a method for prioritizing and interpreting bridge and tunnel group maintenance, provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of a retrieval enhancement large model evaluation criterion and candidate ranking interpretation generation method provided by an embodiment of the present invention; Figure 3 This is a block diagram of a bridge and tunnel group maintenance priority ranking and interpretation generation device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a bridge and tunnel group maintenance priority ranking and interpretation generation device provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0023] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0024] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0025] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0026] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0027] Preferably, annual status data of bridges and tunnels is collected, and data aggregation, standardization, record merging, missing field marking, and profile generation are performed to obtain an annual status profile module corresponding to each bridge or tunnel. An annual status profile is then generated based on this module. The annual status profile includes asset identifiers, annual identifiers, status fields, tag fields, missing fields, source citations, and confidence level markers, including: Collect annual status data for bridges and tunnels, which includes bridge and tunnel asset ledgers, inspection and evaluation data, defect records, maintenance records, traffic and road network information, and manually supplemented records. The bridge and tunnel asset ledgers include, but are not limited to, the main index and asset names. Using the asset number as the primary index, the annual status data of the bridges and tunnels are collected. When asset numbers are missing or inconsistent, the asset name, the road where it is located, and the spatial location auxiliary fields are combined to obtain the asset alignment record; The asset alignment records are standardized, merged, and marked with missing fields to obtain structured annual status data for bridges and tunnels. The standardization process includes unifying units and checking for outliers in numerical fields and extracting keywords and tags from text fields. The record merging process includes merging according to time, severity, and data source reliability. The missing field marking process includes explicitly marking the missing status of key fields. Based on structured annual bridge and tunnel status data, tasks are prioritized according to bridge and tunnel maintenance priorities and processed through profile generation to obtain an annual status profile module for each bridge or tunnel, while retaining the original source of key fields. The annual status profile module includes basic attribute units, technical condition units, defect characteristic units, maintenance history units, traffic function units, management concern units, and data integrity units. The annual status profile module is used to generate an annual status profile, which includes asset identifier, annual identifier, status field, tag field, missing field, source reference, and confidence level marker.

[0028] It should be noted that in some embodiments, the annual status data of bridges and tunnels is first organized. This annual status data originates from business documents such as asset ledgers, inspection and evaluation records, and maintenance records. The system uses the asset number as the primary index and cleans, aligns, standardizes, and marks missing data from different sources to create an annual status profile for each bridge or tunnel. The annual status profile records the asset's basic attributes, technical condition, defects, maintenance history, and data integrity information for the target year, while retaining the original source of key fields.

[0029] It should be further explained that the bridge and tunnel annual status profile construction method is used to organize the scattered bridge and tunnel asset data within the target year into a unified structured status record. The annual status profile uses asset number and year identifier as the main index, recording information such as the basic attributes, technical condition, defect characteristics, maintenance history, traffic function, management concerns, and data integrity of each bridge or tunnel in the target year. The annual status profile serves as the data foundation for subsequent bridge and tunnel-level knowledge evidence package generation, retrieval enhancement, large-scale model reasoning, and consistency verification.

[0030] To ensure real-time data updates, the system integrates data sources including bridge and tunnel asset ledgers, inspection and evaluation data, defect records, maintenance records, traffic and road network information, and manually supplemented records. This data can originate from bridge and tunnel management systems, structured inspection reports, and business ledgers. The system uses asset numbers as the primary index for data aggregation. When asset numbers are missing or inconsistent across different data sources, the system matches them using auxiliary fields such as asset name, road location, and spatial location, and writes the matching results to the asset alignment record. After asset alignment, the system standardizes the original fields. For enumerated fields such as asset type, technical condition level, and defect type, the system uses a unified field mapping table for encoding. For numerical fields such as service life, number of maintenance visits, and inspection intervals, the system standardizes units and checks for outliers. For text fields such as defect descriptions, inspection conclusions, and manual remarks, the system extracts keywords and tags related to maintenance priority ranking. The field standardization results are saved synchronously with the original data sources so that subsequent report interpretations can trace the ranking basis. The annual status profile is organized into several status units according to the bridge and tunnel maintenance priority ranking tasks. The basic attribute unit records information such as asset number, asset type, and road location; the technical condition unit records information such as inspection and evaluation results, technical condition level, and abnormal evaluation markers; the defect characteristic unit records information such as major defects, defect locations, and recurring defect identifiers. Status units such as maintenance history, traffic function, management concern, and data integrity store corresponding fields, tags, and source records in the same manner. For the same asset with multiple inspection, defect, or maintenance records within a target year, the system merges them according to time, severity, and data source reliability. Inspection data prioritizes retaining the latest inspection conclusions and the most unfavorable technical condition level within the target year; defect data prioritizes retaining records of major and abnormal defects; maintenance data retains the most recent maintenance time and maintenance record integrity markers. Original records not used during the merging process are retained as source records in the profile reference information. The system explicitly marks missing key fields. When fields such as technical condition level, defect description, or maintenance record are missing, the annual status profile records the missing field name, missing type, and review marker. For missing fields, the system does not generate definitive alternative conclusions. Missing markers are incorporated into the subsequent retrieval enhancement model inference and consistency verification process along with the annual status profile, serving as a constraint on uncertainty in the candidate ranking interpretation.

[0031] Preferably, after the annual status profile is generated, the system outputs a structured profile record. This record includes asset identifier, annual identifier, status field, tag field, missing field, source citation, and confidence level marker. The bridge and tunnel level knowledge evidence package retrieves evidence based on the asset type, defect tags, technical condition, and management concern tags in the profile; the retrieval enhancement big model reasoning method inputs the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, and reasoning constraints into the big language model to generate Rubric evaluation criteria and candidate ranking interpretation results; the consistency verification method checks the consistency between the candidate ranking interpretation results and the original data based on the fields and source citations in the profile.

[0032] Preferably, external evaluation materials and unstructured data sources based on expert guidance are collected, and source records are established and semantic boundaries are segmented to obtain external basis knowledge units. Simultaneously, the unstructured data sources based on expert guidance are converted into structured conditions that can match the fields of the annual status profile to obtain expert guidance units, including: Collect external evaluation materials and unstructured data sources based on expert guidance; External evaluation materials are used to establish source records and semantic boundary blocks to obtain external basis knowledge units. The external evaluation materials include bridge and tunnel inspection and evaluation standards, local management documents and annual management requirements. The external basis knowledge units record the knowledge number, source material, applicable object, triggering condition, knowledge content and confirmation status. Based on unstructured data sources with expert guidance, structured conditions that can be matched with the annual status profile fields are extracted to obtain expert guidance units; the expert guidance units are used to express the manual attention rules and review requirements in the target annual maintenance priority ranking.

[0033] In some embodiments, both expert guidance and external evaluation materials are structured simultaneously. Expert guidance is segmented, element-extracted, and condition-transformed to form expert guidance units, used to express key concerns, priority reviews, and management requirements in the target year's ranking. External evaluation materials are segmented, source-marked, and applicability-extracted to form external basis knowledge units, used to express the role of standards, policies, and management documents in evaluating maintenance priorities. These structured results are entered into the bridge and tunnel maintenance knowledge base and participate in subsequent retrieval and large-scale model reasoning.

[0034] It should be noted that the method for generating structured and version identifiers for expert opinions and external evaluation criteria is used to organize the standards, regulations, policies, and expert guidance materials involved in prioritizing bridge and tunnel maintenance into citationable and verifiable structured criteria, and to generate version identifiers for the evaluation criteria in this prioritization task. These version identifiers record the scope of external criteria used in this prioritization task and serve as version constraints for generating Rubric evaluation criteria and candidate ranking interpretation results in subsequent large-scale retrieval enhancement model reasoning.

[0035] In some embodiments, for each bridge or tunnel, the system retrieves and filters data from the bridge and tunnel maintenance knowledge base based on its annual status profile, expert guidance units, and external basis knowledge units to form a bridge-tunnel level knowledge evidence package. This bridge-tunnel level knowledge evidence package is not the result of the large model generation, but rather an evidence context formed before the large model's reasoning. Each piece of evidence in the evidence package is assigned an evidence number and associated with the source knowledge unit, source material, and applicable conditions. When the large language model generates candidate ranking reasons, it can only reference the evidence number from this bridge-tunnel level knowledge evidence package.

[0036] Preferably, the evaluation basis version identifier is extracted through external basis knowledge units and expert guidance units; a knowledge base is constructed based on the annual status profile, external basis knowledge units, and expert guidance units to obtain a bridge and tunnel maintenance knowledge base and a new evaluation basis version identifier, including: By using external knowledge units and expert guidance units, material number, knowledge unit number, source information and update time are extracted to form version text. Then, the version text is hashed or equivalent unique identifier is calculated to obtain the evaluation basis version identifier. Based on the annual status profile, structured external basis knowledge units, and expert guidance units, a knowledge base is constructed to obtain the bridge and tunnel maintenance knowledge base. Based on the evaluation criteria version identifier, a version change judgment is made to obtain version constraints for subsequent retrieval enhancement big model reasoning. The version change judgment includes: when the external evaluation criteria version identifier has not changed, the retrieval enhancement big model reasoning is continued to be executed under the same criteria version constraints; when the external evaluation criteria materials, unstructured data sources based on expert guidance opinions, or annual management priorities change, a new evaluation criteria version identifier is regenerated, and the new evaluation criteria version identifier is used to re-execute bridge and tunnel level knowledge evidence package retrieval and retrieval enhancement big model reasoning.

[0037] In some embodiments, the bridge and tunnel maintenance knowledge base stores structured external basis knowledge units, expert experience, and maintenance-related knowledge. For each bridge or tunnel, the system retrieves relevant knowledge content from the bridge and tunnel maintenance knowledge base based on its annual condition profile, expert guidance units, and external basis knowledge units, and then filters it to form a bridge-tunnel level knowledge evidence package. Each piece of evidence in the bridge-tunnel level knowledge evidence package has an evidence number and is associated with the source knowledge unit, applicable conditions, and corresponding asset. This evidence package serves as input to the retrieval-enhanced large-scale model reasoning, constraining the model to generate scoring reasons and evidence citations.

[0038] It should be noted that the system first accesses external evaluation materials and expert guidance. External evaluation materials include bridge and tunnel inspection and evaluation standards, local management documents, annual management requirements, and other relevant documents. The system establishes source records for these materials, recording the material name, source unit, applicable object, update time, and confirmation status. For expert guidance data, the system retains the original text of the opinion, its source, and applicable year, and incorporates it into a structured processing workflow. The system segments the external evaluation basis according to the semantic boundaries of the material content. Standards and specifications are segmented according to clauses, chapters, or business requirements; policy documents are segmented according to management requirements, regulatory objects, or scope of application; and expert guidance data is segmented according to individual opinions. Each segment forms an external basis knowledge unit, recording the knowledge number, source material, applicable object, triggering conditions, knowledge content, and confirmation status. After structured processing, expert guidance opinions form expert guidance units, used to express the rules for manual attention and review requirements in the target year's maintenance priority ranking. Opinions focusing on main road bridges and tunnels, assets with recurring defects entering review, and assets with safety hazards having their attention level increased are converted into structured conditions that can be matched with the annual status profile fields. The system generates evaluation basis version identifiers for the confirmed external basis knowledge units and expert guidance units. During generation, the system extracts the material number, knowledge unit number, source information, and update time involved in this ranking task, and forms version text in a fixed order. This version text is then hashed or used to calculate an equivalent unique identifier to obtain the evaluation basis version identifier. This identifier, along with the external basis knowledge units and expert guidance units, enters the subsequent retrieval enhancement model inference process.

[0039] Preferably, based on the applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit, a search is conducted from the bridge and tunnel maintenance knowledge base to obtain a set of candidate evidence; based on screening rules, the set of candidate evidence is screened according to asset applicability, source reliability, and content relevance to obtain a bridge and tunnel level knowledge evidence package, which is then marked as lacking evidence, including: Based on the applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit, a combination of field condition matching, keyword matching, and semantic retrieval is used to retrieve candidate evidence from the bridge and tunnel maintenance knowledge base to obtain a set of candidate evidence; the applicable conditions include asset type, technical condition, defect labels, traffic function, and management concern labels. Based on the screening rules, the candidate evidence set is screened according to asset applicability, source reliability and content relevance to obtain the evidence set; the screening rules include retaining the candidate evidence with clear source, clear applicable conditions and a matching degree with the annual status profile higher than a preset threshold for content-related candidate evidence, and retaining the missing evidence mark for cases where no valid evidence is found. For each piece of evidence in the evidence set, an evidence number, associated source knowledge unit, and source material are assigned. The bridge and tunnel level knowledge evidence package is saved together with the asset number and evaluation basis version identifier to obtain the bridge and tunnel level knowledge evidence package. Based on the fields and source citations in the annual status profile, the consistency between the bridge and tunnel level knowledge evidence package and the original data is verified. The missing evidence marker is used to constrain the uncertainty prompts in the subsequent candidate ranking interpretation.

[0040] In some embodiments, the evaluation basis version identifier is used to distinguish the impact of "asset status changes" and "evaluation basis changes" on the ranking results. When the annual status data of bridges and tunnels is updated but the external evaluation basis version identifier remains unchanged, subsequent large-scale model inference still generates Rubric evaluation criteria and candidate ranking interpretation results under the same basis version constraint. When the external evaluation basis materials, expert guidance opinions, or annual management priorities change, the system regenerates the evaluation basis version identifier and re-executes the bridge and tunnel level knowledge evidence package retrieval and retrieval-enhanced large-scale model inference under the new basis version constraint. The structured external basis knowledge units enter the bridge and tunnel maintenance knowledge base, and the expert guidance units and evaluation basis version identifiers jointly enter the subsequent inference, verification, and report traceability processes.

[0041] Preferably, a fixed prompt template is used to organize the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier, and inference constraints to obtain the input content of the retrieval enhancement model; the input content of the retrieval enhancement model is then input into the retrieval enhancement model to obtain the output content, which includes the asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking interpretation results, cited evidence number, and manual review mark output, including: Using a fixed prompt template, the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier and reasoning constraints are organized to obtain the input content of the retrieval enhancement model. The fixed prompt template organizes the content by using a single bridge or tunnel as the processing unit. The input content of the retrieval enhancement model is input into the retrieval enhancement model to obtain the output content of the retrieval enhancement model. The output content includes the first part, Rubric evaluation criteria, and the second part, candidate ranking interpretation results. The first part, Rubric evaluation criteria, is constrained by external basis knowledge units, expert guidance units, and evaluation basis version identifiers. The second part, candidate ranking interpretation results, includes candidate scores, candidate priorities, ranking reasons, risk triggering factors, evidence number citations, and uncertainty prompts. The second part, candidate ranking interpretation results, are obtained based on the annual status profile and bridge and tunnel level knowledge evidence package. Using a fixed JSON structure, the output of the retrieval enhancement model is saved, resulting in an output that includes asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking interpretation results, cited evidence number, and manual review mark.

[0042] In some embodiments, the system inputs the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier, and inference constraints into a large language model. The large language model then generates Rubric evaluation criteria and candidate ranking interpretation results under the same basis version and the same evidence context. The Rubric (graded scoring) evaluation criteria are used to provide the evaluation dimensions, dimension weights, and scoring anchors for this maintenance priority ranking; the candidate ranking interpretation results are used to provide candidate scores, candidate priorities, ranking reasons, risk triggering factors, evidence citations, and uncertainty prompts.

[0043] It should be noted that, as Figure 2As shown, the retrieval enhancement model's reasoning stage focuses on bridge and tunnel assets. The system inputs annual status profiles, expert guidance units, external evidence knowledge units, bridge and tunnel-level knowledge evidence packages, and reasoning constraints into the large language model. Under these input constraints, the large language model generates Rubric evaluation criteria and candidate ranking interpretation results. The Rubric evaluation criteria express the evaluation dimensions, weights, and scoring anchors used in this ranking; the candidate ranking interpretation results express the candidate scores, candidate priorities, ranking reasons, risk triggering factors, evidence number citations, and uncertainty warnings for a single bridge or tunnel. After the candidate ranking interpretation results are generated, the system performs a rule-based consistency check. The check includes the Rubric evaluation criterion format, scoring range, status profile fact citations, evidence number citations, and expert guidance unit triggering relationships. For candidate content inconsistent with the annual status profile, bridge and tunnel-level knowledge evidence packages, or expert guidance units, the system records an anomaly and generates a review flag. For scores and reasons that pass the check, the system enters the priority correction and comprehensive ranking process. Priority correction performs rule-based processing on the candidate ranking results based on the check results, expert guidance units, and risk triggering conditions. The corrected results form the final maintenance priority ranking table, and an interpretation report is generated simultaneously. The interpretation report records the ranking results, Rubric evaluation criteria, main ranking basis, evidence number citations, uncertainty warnings, and manual review items. The manual review results can be fed back to the expert guidance unit, external basis knowledge unit, and bridge and tunnel maintenance knowledge base for subsequent annual ranking task updates.

[0044] It should be further explained that the bridge and tunnel level knowledge evidence package is an evidence collection formed for a single bridge or tunnel. Based on information such as asset type, technical condition, defect tags, traffic function, and management concern tags in the annual status profile, and combined with the applicable conditions in expert guidance units and external basis knowledge units, the system retrieves relevant knowledge content from the bridge and tunnel maintenance knowledge base. The retrieval process uses a combination of field condition matching, keyword matching, and semantic retrieval to form a candidate evidence set. After the candidate evidence is formed, the system filters it according to asset applicability, source reliability, and content relevance. Content that does not match the current bridge and tunnel asset status, has an unconfirmed source, or is irrelevant to the maintenance priority ranking task is not included in the evidence package. For candidate evidence with similar content, the system prioritizes evidence with a clear source, clear applicable conditions, and a high degree of matching with the annual status profile. In cases where no valid evidence is found, a missing evidence mark is retained in the evidence package. The generated bridge and tunnel level knowledge evidence package is saved along with the asset number and evaluation basis version identifier. Each piece of evidence in the evidence package is assigned an evidence number and associated with the source knowledge unit and source material.

[0045] It should be further explained that the bridge and tunnel domain-specific retrieval enhancement large model (RAG-LLM) adopted can be a general large model. It does not require full parameter post-training optimization. It can complete the domain alignment and adaptation optimization of the annual status evaluation scenario for highway bridge and tunnel assets by only LoRA fine-tuning or QLoRA fine-tuning, thus meeting the needs of scenario application and patent implementation.

[0046] The fine-tuning training corpus is a specially constructed professional corpus for bridge and tunnel operation and maintenance, such as national and industry-level standard provisions like the "Technical Condition Assessment Standard for Highway Bridges" and the "Technical Specification for Highway Tunnel Maintenance"; it can also include professional rule datasets for bridge and tunnel defect risk assessment, priority ranking, and scoring criteria.

[0047] Preferably, the output of the retrieval enhancement model, external basis knowledge units, and expert guidance units are used to perform consistency verification on the output of the retrieval enhancement model, obtaining the verified Rubric evaluation criteria, Rubric dimension weights, and verification marks; finally, a comprehensive score is calculated to form a bridge and tunnel maintenance priority ranking table, including: The Rubric evaluation criteria of the retrieval enhancement model's output content, external basis knowledge units, and expert guidance units are used to perform consistency verification on the output content of the retrieval enhancement model. This results in Rubric evaluation criteria, Rubric dimension weights, and verification marks that pass the verification. The consistency verification includes checking whether the evaluation dimensions, dimension weights, scoring anchors, and external basis references meet the ranking task requirements. When evaluation dimensions are missing, weights do not meet the calculation requirements, scoring anchors cannot correspond to engineering meanings, or external basis references cannot be traced back to structured knowledge units, the corresponding Rubric content is marked as a criterion anomaly. Based on the annual status profile, bridge and tunnel level knowledge evidence package, and expert guidance unit, the candidate ranking interpretation results are verified by data and evidence to obtain the candidate ranking interpretation results that pass the verification. The data and evidence verification includes verifying the scoring reasons, the cited evidence numbers, and the triggering of the expert guidance unit. When a candidate result has external input information, no evidence number, or the evidence does not match the reasons, the anomaly type is recorded and it is restricted from entering the final ranking interpretation. Based on the validated Rubric evaluation criteria, Rubric dimension weights, and review markers, a comprehensive score is calculated, and bridge and tunnel assets that trigger expert attention, have abnormal risks, or lack key data are corrected to obtain the corrected results; the correction includes retaining manual review markers, retaining records of score anomalies, and hiding untraceable explanatory content; Based on the corrected results, the bridge and tunnel maintenance priorities are ranked to obtain a bridge and tunnel maintenance priority ranking table; the bridge and tunnel maintenance priority ranking table includes asset number, comprehensive score, priority level, ranking position and review status.

[0048] In some embodiments, for the candidate ranking interpretation results output by the large language model, the system uses the annual status profile, expert guidance unit, external basis knowledge unit, and bridge-tunnel level knowledge evidence package as verification criteria to check whether the ranking reasons correspond to the bridge-tunnel status facts, whether the cited evidence number exists in the current bridge-tunnel level knowledge evidence package, and whether the expert guidance conditions are responded to in the candidate results. For content that cannot correspond to the status facts, has a non-existent evidence number, does not match the evidence and reasons, or is not responded to by the expert guidance conditions, the system will not use it as the final interpretation basis and will mark the corresponding asset or corresponding dimension as requiring review.

[0049] It should be noted that the system first performs a consistency check on the Rubric evaluation criteria. This check includes verifying whether the evaluation dimensions, dimension weights, scoring anchors, and external references meet the requirements of this ranking task. If an evaluation dimension is missing, the weights do not meet the calculation requirements, the scoring anchors do not correspond to the engineering meaning, or the external references cannot be traced back to the structured knowledge unit, the system will mark the corresponding Rubric content as a criterion anomaly and initiate a manual review or regeneration process. Subsequently, the system performs data and evidence verification on the candidate ranking interpretation results. The scoring reasons in the candidate results should have corresponding factual basis that can be found in the annual status profile, the cited evidence numbers should exist in the bridge and tunnel level knowledge evidence package of the current bridge and tunnel assets, and when an expert guidance unit is triggered, the corresponding attention or review mark should be reflected in the candidate results. For candidate results with issues such as inputting external information, missing evidence numbers, or mismatched evidence and reasons, the system records the anomaly type and restricts their entry into the final ranking interpretation.

[0050] After completing the consistency verification, the system calculates a comprehensive score based on the dimensions that passed verification, the Rubric dimension weights, and the review flags. For bridge and tunnel assets that trigger expert attention, exhibit abnormal risks, or have missing key data, the system retains a manual review flag according to preset rules. For dimensions with insufficient evidence or reasons that fail verification, the system retains a score anomaly record and hides untraceable explanations in the final report. The corrected results form a bridge and tunnel maintenance priority ranking table, including asset number, comprehensive score, priority level, ranking position, and review status.

[0051] The above is an introduction to the method embodiments. The following describes the solution described in this application through device embodiments.

[0052] Figure 3 This is a block diagram illustrating a bridge and tunnel group maintenance priority ranking and interpretation generation apparatus according to an exemplary embodiment. The apparatus is used in a bridge and tunnel group maintenance priority ranking and interpretation generation method. (Refer to...) Figure 3The device includes an annual status profile module 310, an external basis module 320, a knowledge base module 330, a candidate evidence module 340, a large model module 350, and a priority ranking module 360.

[0053] Annual Status Profile Module 310: Used to collect annual status data of bridges and tunnels, perform data aggregation, standardization processing, record merging, missing field marking and profile generation processing to obtain the annual status profile module corresponding to each bridge or tunnel, and generate an annual status profile based on the annual status profile module; the annual status profile includes asset identifier, annual identifier, status field, tag field, missing field, source reference and confidence mark; External basis module 320: used to collect external evaluation basis materials and unstructured data sources based on expert guidance opinions, establish source records and semantic boundary blocks to obtain external basis knowledge units; at the same time, convert the unstructured data sources based on expert guidance opinions into structured conditions that can match the fields of the annual status profile to obtain expert guidance units; Knowledge Base Module 330: Used to extract evaluation basis version identifiers through external basis knowledge units and expert guidance units; and to construct a knowledge base based on annual status profiles, external basis knowledge units, and expert guidance units to obtain a bridge and tunnel maintenance knowledge base and new evaluation basis version identifiers. Candidate Evidence Module 340: This module is used to retrieve candidate evidence from the bridge and tunnel maintenance knowledge base based on the applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit. Based on the screening rules, the candidate evidence is screened according to asset applicability, source reliability, and content relevance to obtain a bridge and tunnel level knowledge evidence package, and missing evidence is marked. Large Model Module 350: This module organizes the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier, and inference constraints using a fixed prompt template to obtain the input content of the retrieval enhancement large model; the input content of the retrieval enhancement large model is then input into the retrieval enhancement large model to obtain the output content, which includes the asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking interpretation results, cited evidence number, and manual review mark output. Priority ranking module 360: It is used to perform consistency verification on the output of the retrieval enhancement model by using the output content of the retrieval enhancement model, external basis knowledge units and expert guidance units, to obtain the Rubric evaluation criteria, Rubric dimension weights and verification marks that pass the verification; finally, it performs comprehensive score calculation to form a bridge and tunnel maintenance priority ranking table.

[0054] A bridge and tunnel group maintenance priority ranking and interpretation generation device, comprising: a processor; and a memory storing computer-readable instructions, wherein when the processor executes the computer-readable instructions, it implements the method described in any one of the above-described bridge and tunnel group maintenance priority ranking and interpretation generation methods.

[0055] Figure 4 This is a schematic diagram of a bridge and tunnel group maintenance priority ranking and interpretation generation device provided in an embodiment of the present invention, as shown below. Figure 4 As shown, the bridge and tunnel group maintenance priority ranking and interpretation generation equipment may include the above-mentioned Figure 3 The bridge and tunnel group maintenance priority ranking and interpretation generation device shown. Optionally, the bridge and tunnel group maintenance priority ranking and interpretation generation device 410 may include a first processor 2001.

[0056] Optionally, the bridge and tunnel group maintenance priority ranking and interpretation generation device 410 may also include a memory 2002 and a transceiver 2003.

[0057] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0058] The following is combined with Figure 4 A detailed description of each component of the bridge and tunnel group maintenance priority ranking and interpretation generation equipment 410 is provided below: The first processor 2001 is the control center of the bridge and tunnel group maintenance priority ranking and interpretation generation device 410. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0059] Optionally, the first processor 2001 can perform various functions of the bridge and tunnel group maintenance priority sorting and interpretation generation device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0060] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 are shown in the diagram.

[0061] In a specific implementation, as one example, the bridge and tunnel group maintenance priority ranking and interpretation generation device 410 may also include multiple processors, for example... Figure 4 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0062] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0063] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently, and may be connected via the interface circuit of the bridge and tunnel maintenance priority sorting and interpretation generation device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0064] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0065] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 4(Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0066] Optionally, the transceiver 2003 can be integrated with the first processor 2001 or exist independently, and can be connected to the interface circuit of the bridge and tunnel group maintenance priority sorting and interpretation generation device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0067] It should be noted that, Figure 4 The structure of the bridge and tunnel group maintenance priority ranking and interpretation generation device 410 shown in the figure does not constitute a limitation on the router. The actual knowledge structure identification device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0068] Furthermore, the technical effects of the bridge and tunnel group maintenance priority ranking and interpretation generation device 410 can be referred to the technical effects of the bridge and tunnel group maintenance priority ranking and interpretation generation method described in the above method embodiments, and will not be repeated here.

[0069] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0070] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0071] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0072] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0073] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0074] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0075] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0076] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0077] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0078] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0079] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0080] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0081] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for prioritizing and interpreting bridge and tunnel group maintenance, characterized in that, The method includes: S1: Collect annual status data of bridges and tunnels, perform data aggregation, standardization processing, record merging, missing field marking and profile generation processing to obtain the annual status profile module corresponding to each bridge or tunnel, and generate an annual status profile based on the annual status profile module; the annual status profile includes asset identifier, annual identifier, status field, tag field, missing field, source reference and confidence mark; S2: Collect external evaluation materials and unstructured data sources based on expert guidance, establish source records and segment semantic boundaries to obtain external basis knowledge units; at the same time, convert the unstructured data sources based on expert guidance into structured conditions that can match the fields of the annual status profile to obtain expert guidance units. S3: Extract the evaluation basis version identifier through external basis knowledge units and expert guidance units; construct a knowledge base based on the annual status profile, external basis knowledge units and expert guidance units to obtain the bridge and tunnel maintenance knowledge base and the new evaluation basis version identifier; S4: Based on the applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit, retrieve candidate evidence from the bridge and tunnel maintenance knowledge base to obtain a set of candidate evidence; based on the screening rules, screen the candidate evidence set according to asset applicability, source reliability, and content relevance to obtain a bridge and tunnel level knowledge evidence package, and mark it as missing evidence. S5: Using a fixed prompt template, organize the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier, and inference constraints to obtain the input content of the retrieval enhancement model; input the input content of the retrieval enhancement model into the retrieval enhancement model to obtain the output content, which includes the asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking interpretation results, cited evidence number, and manual review mark output; S6: Using the output content of the retrieval enhancement model, external basis knowledge units, and expert guidance units, the consistency of the output of the retrieval enhancement model is verified to obtain the Rubric evaluation criteria, Rubric dimension weights, and verification marks that pass the verification; finally, a comprehensive score is calculated to form a bridge and tunnel maintenance priority ranking table.

2. The method for prioritizing and interpreting bridge and tunnel group maintenance according to claim 1, characterized in that, The S1 collects annual status data of bridges and tunnels, performs data aggregation, standardization processing, record merging, missing field marking and profile generation processing to obtain an annual status profile module corresponding to each bridge or tunnel, and generates an annual status profile based on the annual status profile module. The annual status profile includes asset identifier, annual identifier, status field, tag field, missing field, source reference, and confidence level marker, including: S11: Collect annual status data for bridges and tunnels, which includes bridge and tunnel asset ledgers, inspection and evaluation data, defect records, maintenance records, traffic and road network information, and manually supplemented records; S12: Using the asset number as the primary index, collect the annual status data of the bridge and tunnel; S13: When asset numbers are missing or inconsistent, the asset name, the road where it is located, and the spatial location auxiliary fields are combined for matching to obtain asset alignment records; S14: Standardize the asset alignment records, merge the records, and mark missing fields to obtain structured bridge and tunnel annual status data; S15: Based on the structured annual status data of bridges and tunnels, the tasks are sorted according to the priority of bridge and tunnel maintenance and processed through portrait generation to obtain the annual status portrait module corresponding to each bridge or tunnel, and the original source corresponding to the key fields is retained. The annual status portrait includes asset identifier, annual identifier, status field, tag field, missing field, source reference and confidence mark.

3. The method for prioritizing and interpreting bridge and tunnel group maintenance according to claim 1, characterized in that, The S2 process involves collecting external evaluation materials and unstructured data sources based on expert guidance, establishing source records, and segmenting semantic boundaries to obtain external basis knowledge units. Simultaneously, the unstructured data source based on expert guidance is converted into structured conditions that can match the fields of the annual status profile, resulting in an expert guidance unit, including: S21: Collect external evaluation materials and unstructured data sources based on expert guidance; S22: Using external evaluation materials, source records are established and semantic boundaries are segmented to obtain external basis knowledge units; S23: Based on the unstructured data source with expert guidance, extract the structured conditions that can be matched with the annual status profile fields to obtain the expert guidance unit.

4. The method for prioritizing and interpreting bridge and tunnel group maintenance according to claim 1, characterized in that, S3 extracts the evaluation basis version identifier through external knowledge units and expert guidance units; A knowledge base was constructed based on annual status profiles, externally referenced knowledge units, and expert guidance units, resulting in a bridge and tunnel maintenance knowledge base and a new evaluation basis version identifier, including: S31: By using external knowledge units and expert guidance units, extract the material number, knowledge unit number, source information and update time to form a version text, and perform hash calculation or equivalent unique identifier calculation on the version text to obtain the evaluation basis version identifier; S32: Based on the annual status profile, structured external basis knowledge units, and expert guidance units, a knowledge base is constructed to obtain the bridge and tunnel maintenance knowledge base; S33: Based on the evaluation criteria version identifier, determine the version change and obtain the version constraints used for subsequent retrieval enhancement of the large model inference.

5. The method for prioritizing and interpreting bridge and tunnel group maintenance according to claim 1, characterized in that, S4 retrieves candidate evidence from the bridge and tunnel maintenance knowledge base based on the applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit. Based on the screening rules, the candidate evidence set was screened according to asset applicability, source reliability, and content relevance to obtain a bridge and tunnel level knowledge evidence package, which was then marked as lacking evidence, including: S41: Based on the applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit, a combination of field condition matching, keyword matching, and semantic retrieval is used to retrieve candidate evidence from the bridge and tunnel maintenance knowledge base. S42: Based on the screening rules, the candidate evidence set is screened according to asset applicability, source reliability, and content relevance to obtain the evidence set; S43: Set an evidence number, associated source knowledge unit, and source material for each piece of evidence in the evidence set, and save the bridge and tunnel level knowledge evidence package together with the asset number and evaluation basis version identifier to obtain the bridge and tunnel level knowledge evidence package; S44: Verify the consistency between the bridge and tunnel level knowledge evidence package and the original data based on the fields and source references in the annual status profile.

6. The method for prioritizing and interpreting bridge and tunnel group maintenance according to claim 1, characterized in that, The S5 uses a fixed prompt template to organize the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier and reasoning constraints to obtain the input content of the retrieval enhancement model; The input content of the retrieval enhancement model is input into the retrieval enhancement model to obtain the output content. The output content includes asset number, evaluation basis version identifier, Rubric evaluation criterion, candidate ranking interpretation result, cited evidence number, and manual review mark output, including: S51: Using a fixed prompt template, the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier and reasoning constraints are organized to obtain the input content of the retrieval enhancement model. The fixed prompt template organizes the content by using a single bridge or tunnel as the processing unit. S52: Input the input content of the retrieval enhancement model into the retrieval enhancement model to obtain the output content of the retrieval enhancement model; the output content includes the first part of the content Rubric evaluation criteria and the second part of the content candidate ranking interpretation results. The first part of the content Rubric evaluation criteria is constrained by external basis knowledge units, expert guidance units, and evaluation basis version identifiers. The second part of the content candidate ranking interpretation results includes candidate scores, candidate priorities, ranking reasons, risk triggering factors, evidence number citations, and uncertainty prompts. The second part of the content candidate ranking interpretation results is obtained based on the annual status profile and bridge and tunnel level knowledge evidence package. S53: Using a fixed JSON structure, the output content of the retrieval enhancement model is saved, resulting in an output including asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking interpretation results, cited evidence number, and manual review mark.

7. The method for prioritizing and interpreting bridge and tunnel group maintenance according to claim 1, characterized in that, S6 uses the output content of the retrieval enhancement model, external basis knowledge unit and expert guidance unit to perform consistency verification on the output of the retrieval enhancement model, and obtains the Rubric evaluation criteria, Rubric dimension weights and verification marks that pass the verification. Finally, a comprehensive score is calculated to generate a priority ranking table for bridge and tunnel maintenance, including: S61: Using the output content of the retrieval enhancement model, external basis knowledge units, and expert guidance units, the Rubric evaluation criteria of the output content of the retrieval enhancement model are validated for consistency, and the validated Rubric evaluation criteria, Rubric dimension weights, and verification marks are obtained. S62: Based on the annual status profile, bridge and tunnel level knowledge evidence package and expert guidance unit, perform data and evidence verification on the candidate ranking interpretation results to obtain the candidate ranking interpretation results that pass the verification; S63: Calculate the comprehensive score based on the verified Rubric evaluation criteria, Rubric dimension weights, and review markers, and correct bridge and tunnel assets that trigger expert attention, have abnormal risks, or lack key data to obtain the corrected results; S64: Based on the corrected results, prioritize bridge and tunnel maintenance to obtain a bridge and tunnel maintenance priority ranking table; the bridge and tunnel maintenance priority ranking table includes asset number, comprehensive score, priority level, ranking position and review status.

8. A bridge and tunnel group maintenance priority ranking and interpretation generation device, wherein the bridge and tunnel group maintenance priority ranking and interpretation generation device is used to implement the bridge and tunnel group maintenance priority ranking and interpretation generation method as described in any one of claims 1-7, characterized in that, The device includes: Annual Status Profile Module: This module is used to collect annual status data of bridges and tunnels, perform data aggregation, standardization, record merging, missing field marking, and profile generation to obtain the annual status profile module corresponding to each bridge or tunnel. An annual status profile is generated based on the annual status profile module. The annual status profile includes asset identifier, annual identifier, status field, tag field, missing field, source reference, and confidence level mark. External basis module: used to collect external evaluation basis materials and unstructured data sources based on expert guidance, establish source records and semantic boundary segmentation to obtain external basis knowledge units; at the same time, convert the unstructured data sources based on expert guidance into structured conditions that can match the fields of the annual status profile to obtain expert guidance units; Knowledge Base Module: Used to extract evaluation basis version identifiers through external basis knowledge units and expert guidance units; based on the annual status profile, external basis knowledge units, and expert guidance units, a knowledge base is constructed to obtain a bridge and tunnel maintenance knowledge base and new evaluation basis version identifiers; Candidate Evidence Module: This module retrieves candidate evidence from the bridge and tunnel maintenance knowledge base based on applicable conditions in the annual status profile, expert guidance unit, and external basis knowledge unit. Based on screening rules, the candidate evidence set is filtered according to asset applicability, source reliability, and content relevance to obtain a bridge and tunnel level knowledge evidence package, and missing evidence is marked. Large Model Module: This module organizes the annual status profile, expert guidance unit, external basis knowledge unit, bridge and tunnel level knowledge evidence package, evaluation basis version identifier, and inference constraints using a fixed prompt template to obtain the input content for the retrieval enhancement large model. The input content is then input into the retrieval enhancement large model to obtain the output content, which includes the asset number, evaluation basis version identifier, Rubric evaluation criteria, candidate ranking explanation results, cited evidence number, and manual review mark. Priority ranking module: It is used to perform consistency verification on the output of the retrieval enhancement model by using the output content of the retrieval enhancement model, external basis knowledge units and expert guidance units, to obtain the Rubric evaluation criteria, Rubric dimension weights and verification marks that pass the verification; finally, it performs comprehensive score calculation to form a bridge and tunnel maintenance priority ranking table.

9. A device for prioritizing and interpreting bridge and tunnel group maintenance, characterized in that, The bridge and tunnel group maintenance priority ranking and interpretation generation device includes: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the method as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 7.

Images