A method, system and medium for automatically checking physical consistency of hydro-meteorological engineering report parameters

Abstract

The present application relates to a kind of hydrology meteorological engineering report parameter physical consistency automatic examination method, system and medium, belong to engineering information and intelligent examination technical field.The method obtains the report to be examined, parses text, table, figure note and appendix, identifies engineering type, design stage, site condition, elevation datum and applicable specification;Accordingly, dynamically generate parameter examination requirement library, extract wind speed, tidal level, wave and other parameters to form report parameter list with multi-dimensional label;According to physical constraint model, candidate relationship group is constructed, and label matching of working condition, return period, site and specification is carried out, and fixed formula is called during matching to review calculation, and it is transferred into manual review if it is not matched;According to deviation, update verification state, and error level parameter is prohibited to be stored in library.The present application solves the problem of cross-working condition false alarm and generates model numerical illusion by pre-label verification and deterministic calculation, and improves the examination accuracy.

Description

Technical Field

[0001] This invention belongs to the field of engineering informatization and intelligent review technology, specifically relating to a computer-implemented automatic review method, system and medium for physical consistency of parameters in hydrological and meteorological engineering reports. Background Technology

[0002] Hydrometeorological engineering reports are widely used in offshore wind power, marine photovoltaic projects, port engineering, marine engineering, water conservancy projects, power engineering, disaster prevention and mitigation projects, and related infrastructure construction. These reports typically include project overview, meteorological conditions, tide levels, water levels, waves, current velocity, wind speed, wind pressure, precipitation, evaporation, thunderstorms, return period, elevation datum, statistical time period, design conditions, regulatory basis, and calculation sheets. The parameters in the report are crucial for design, construction, verification, and operation management, and also serve as data sources for subsequent structural calculations, load combinations, site selection, and project archiving.

[0003] In actual review and verification, errors in hydrological and meteorological reports do not always manifest as typos, incorrect table of contents numbering, or inconsistencies in the same field. More common and subtle errors lie in inconsistencies in engineering constraints between different physical quantities. For example, design wind speed and design wind pressure should meet certain conversion relationships under the same site, operating conditions, return period, and statistical time period; design tide level, water level, and elevation benchmarks should be consistent or convertible; wave height, period, wave direction, and water depth should meet the corresponding sea state and applicable specification conditions; and when the same parameter appears in the main text, tables, figure captions, appendices, and calculation sheets, its source, unit, return period, and applicable conditions should correspond to each other.

[0004] Existing automated report review technologies typically employ keyword matching, template comparison, rule lists, consistency checks for fields with the same name, or semantic review using large language models. While these methods can identify formatting, description, numbering, and some explicit data issues, they often lack verifiable deterministic calculation links for implicit engineering relationships such as wind speed-wind pressure, water level-benchmark, wave parameters, tidal conditions, and consistency across different chapters. Particularly when reports contain multiple return periods, multiple elevation benchmarks, multiple sites, or multiple operating conditions, a system that relies solely on keywords to identify two values ​​and directly compares them is prone to generating false alarms related to cross-return periods, cross-benchmarks, or cross-operating conditions.

[0005] Furthermore, while enhanced retrieval generation and large language models are being used in engineering report review, and these models can effectively understand context and generate natural language review comments, they are not suitable for solely assessing engineering numerical values. Directly judging the accuracy of wind pressure calculations using generative models can easily lead to problems such as incorrect formula selection, omissions in unit conversions, neglect of return period conditions, inconsistent rounding, and conclusion drift. Even repeated reviews of the same report may yield different results due to differences in prompts, context windows, or model sampling. Such instability can affect the traceability of review conclusions and the boundaries of engineering responsibility.

[0006] Therefore, it is necessary to provide a technical solution that combines engineering information identification, dynamic generation of parameter requirements, report parameter list, label matching judgment, fixed formula calculation verification, status stop, and fixed basis for review generation. This solution should be able to determine whether parameters belong to the same comparable relation group before generating review opinions, and complete the calculation by a deterministic module when matching occurs. If there is no match, the calculation verification should be stopped and a manual review should be prompted, thereby improving the accuracy, interpretability, and engineering usability of the report review. Summary of the Invention

[0007] The purpose of this invention is to provide an automatic review method, system, and medium for physical consistency of parameters in hydrological and meteorological engineering reports. This invention addresses the problem that existing automatic review schemes mainly rely on keywords, templates, or human experience, making it difficult to detect inconsistencies between wind speed and wind pressure, water level and elevation benchmarks, and wave parameters and operating conditions in a timely manner. It also avoids situations where incorrect formula selection, unit omission, or inconsistent conclusions may occur when generating models and making numerical judgments directly.

[0008] To achieve the above objectives, the technical solution of the present invention is: an automatic method for reviewing the physical consistency of parameters in hydrological and meteorological engineering reports, wherein a computer system performs the following data processing steps:

[0009] Obtain the hydrological and meteorological engineering report to be reviewed, and use a document parsing engine to perform structured parsing on the main text, tables, figure captions, appendices and calculation sheets in the report to obtain structured text content, numerical content and parameter source location information;

[0010] Based on the structured text content, a natural language processing model is used to identify engineering information, which includes at least the project type, design stage, site conditions, sea area location, elevation datum, and applicable standards.

[0011] Based on the project information, a parameter review requirement library is dynamically generated through a rule engine. The parameter review requirement library includes the hydrological and meteorological parameters that should be provided for the corresponding type of project, the corresponding applicable condition fields, and the source of the requirements.

[0012] Hydrological and meteorological engineering parameters are extracted from the structured text content and numerical content using an information extraction model to form a structured report parameter list. Each parameter item in the report parameter list includes at least the parameter type, parameter value, unit, source location, operating condition label, return period label, site label, benchmark label, specification version label, and verification status.

[0013] Based on a pre-defined physical constraint model library and parameter type mapping table, candidate physical relationship groups are constructed from the report parameter list according to the pre-defined physical relationships between parameters;

[0014] The candidate physical relationship group is subjected to multi-dimensional label matching judgment through a matching judgment matrix. The multi-dimensional labels include operating conditions, return period, site, benchmark, specification version, and statistical period.

[0015] When the candidate physical relationship group meets the multidimensional label matching conditions, the corresponding deterministic physical constraint model is called to perform calculation verification, obtain the physical calculation value and the deviation value, and save the calculation result as a locked state;

[0016] When the candidate physical relationship group does not meet the multidimensional label matching conditions, the physical formula calculation verification is not performed, a label mismatch prompt is generated, and the relevant parameter items are set to a state pending manual verification.

[0017] The verification status in the report parameter list is updated according to the deviation value and the preset threshold rule, and the parameter items that reach the preset error level are set to the unpassable status.

[0018] The physical calculation value, deviation value, source location, and verification status are output as review results to the review interface.

[0019] Furthermore, the process of using a natural language processing model to identify engineering information includes: extracting information such as project type, design stage, province, marine or land-based site conditions, marine location, elevation datum, statistical period, and standard number and name cited in the report from the report name, project overview, technical basis section, design parameter section, table titles, figure captions, appendix names, and parameter context using named entity recognition and text classification models. The identification results are then used as inputs to generate a parameter review requirement library and matching judgment rules.

[0020] Furthermore, the parameter review requirement library is dynamically generated based on engineering information; for marine photovoltaic, offshore wind power, port or marine engineering, the parameter review requirement library shall include at least one or more of the following: design wind parameters, wind pressure and conversion basis, tide level or design water level, design waves, design current velocity, applicable specifications and versions, and joint working condition description, and configure necessary fields, applicable conditions and missing item prompts for each parameter.

[0021] Furthermore, the parameter items in the report parameter list also include original text excerpts, chapters or page numbers, table rows and columns, statistical time periods, extraction confidence levels, and relationship group identifiers. The relationship group identifiers are used to group different physical quantities from the same site, the same working condition, the same return period, the same elevation datum, or the same original text context into the same candidate physical relationship group.

[0022] Furthermore, the multidimensional label matching judgment is implemented through a matching judgment matrix, rule base, key-value mapping or database table; when any necessary label in the working condition, recurrence period, site, benchmark, specification version or statistical period is missing or conflicting, the physical formula calculation review is not performed, a label mismatch prompt is generated and the relevant parameter items are set to a state pending manual review.

[0023] Furthermore, the deterministic physical constraint model includes a wind speed-wind pressure physical constraint model, which is based on W=V 2 / K is used to calculate wind pressure, where W is wind pressure, V is wind speed, and K is a conversion factor related to the unit system, air density, statistical period, or standard version.

[0024] Furthermore, the deterministic physical constraint model also includes one or more of the following: wave parameter physical constraint model, water level condition physical constraint model, elevation datum conversion constraint model, tide level applicability condition constraint model, and cross-chapter consistency constraint model for parameters of the same origin.

[0025] Furthermore, the verification status includes pending verification, verified, prompt, calculation error, missing item or pending confirmation, pending manual review, and unpassable; when a parameter item is set to an unpassable status, the system controls the corresponding parameter item to not enter the report review approval list, structured results library, historical case positive example library, or subsequent automatic calculation module.

[0026] This invention also provides an automatic review system for the physical consistency of parameters in hydrological and meteorological engineering reports, which performs the method described above, including:

[0027] The document parsing module is used to obtain the hydrological and meteorological engineering report to be reviewed and parse the main text, tables, figure captions, appendices and calculation sheets to obtain the text content, numerical content and parameter source location information;

[0028] The engineering information recognition module is used to recognize engineering information based on the parsed text content;

[0029] The parameter review requirement library generation module is used to generate a parameter review requirement library based on the project information.

[0030] The report parameter list generation module is used to extract hydrological and meteorological engineering parameters from text and numerical content and generate a report parameter list.

[0031] The candidate physics relation group construction module is used to construct candidate physics relation groups from the report parameter list according to the preset physics constraint model;

[0032] The multidimensional label matching and judgment module is used to perform multidimensional label matching and judgment on candidate physical relationship groups;

[0033] The fixed formula calculation verification module is used to call the deterministic physical constraint model for calculation verification under the condition that the label matching is satisfied;

[0034] The status update and process control module is used to update the verification status according to the deviation value and threshold rules, and set the parameter items that reach the error level to the unpassable status;

[0035] The review results generation module is used to output physical calculation values, deviation values, source locations, and verification status.

[0036] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0037] It should be particularly noted that the technical contribution of this invention lies not in the hydrological and meteorological review rules themselves, but in how to reliably perform physical consistency verification through improvements in the data processing architecture of the computer system. Specifically: (1) By using a multi-dimensional label matching judgment matrix, the system automatically identifies whether parameters belong to the same comparable engineering conditions before calculation, thus solving the technical problems of false alarms across operating conditions, across recurrence periods, and across benchmarks from the perspective of data processing flow; (2) By saving the calculation results of the deterministic physical constraint model as a locked state, the system architecture boundary between numerical calculation and natural language generation is clearly defined, thus solving the technical problems of illusions and conclusion drift in numerical judgments caused by large language models; (3) By using a process control mechanism for unpassable states, the system blocks the backward propagation of erroneous parameters at the data flow level. The solution to the above technical problems does not depend on changes in the content of the review rules, but on improvements in the data processing architecture and process control mechanism within the computer system.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] First, this invention can discover hidden engineering contradictions between different physical quantities, and is not limited to comparing fields with the same name and checking keyword rules. For example, the system can group design wind speed and design wind pressure under the same return period into the same relationship group, and use a fixed formula to calculate and verify whether the wind pressure value is reasonable.

[0040] Second, this invention achieves specification-driven review through engineering information identification and a parameter review requirement database. The system first determines the project type, design stage, and site conditions, and then generates the parameter requirements that such projects should possess, thereby identifying problems such as missing report items, incomplete applicable conditions, unclear specification versions, and insufficient calculation basis.

[0041] Third, this invention reduces false alarms across operating conditions by using a multi-dimensional label matching judgment matrix. Before calculation, the system automatically checks the labels for operating condition, return period, site, benchmark, specification version, and statistical period. It only proceeds to formula calculation when all necessary labels are consistent or the conversion relationship can be confirmed. If any label is missing or conflicting, the system does not perform numerical calculation but generates a mismatch prompt and transfers the data to manual review. This avoids erroneous numerical comparisons across return periods, benchmarks, or operating conditions from the perspective of data processing.

[0042] Fourth, this invention, through a mechanism for locking and storing the calculation results of fixed formulas, uses the calculation results of the deterministic physical constraint model as a fixed basis that the review generation module must adopt, thus achieving architectural separation between numerical calculation and natural language generation. The large language model or generation module is only used to express and organize the natural language text of review opinions, while the fixed formula module is used to complete deterministic numerical judgments. The generation module must not recalculate, replace, or reduce the locked numerical conclusions. This architectural separation design fundamentally reduces the technical risks of illusions and conclusion drift arising from the large language model in numerical judgments, and improves the traceability of review conclusions and the clarity of engineering responsibility boundaries.

[0043] Fifth, this invention controls the subsequent data flow by controlling the unpassable state, enabling the review conclusions to influence the entry of results into the database, the pass list, the accumulation of historical cases, and subsequent calculation modules, forming a complete process from problem discovery, source location, report correction to re-review.

[0044] Sixth, this invention retains the source location of parameters, original text fragments, relation group identifiers, and deviation calculation basis, enabling reviewers to quickly locate the original report text and verify the system conclusions, thereby improving the interpretability, traceability, and efficiency of the review process. Attached Figure Description

[0045] Figure 1 A flowchart of an automatic review method for physical consistency of parameters in hydrological and meteorological engineering reports provided in this embodiment of the invention.

[0046] Figure 2 This is the overall output diagram of the system's review of the driving parameters for the execution of hydrological and meteorological reports for a certain project.

[0047] Figure 3 This document provides the first page and review coverage output of the specification-driven parameter review file generated by the system for a specific project.

[0048] Figure 4 The system outputs a parameter review requirement library to identify engineering information for a specific project.

[0049] Figure 5 This is the output of the system to generate a report parameter list for a specific project and to perform a verification calculation using the fixed formula for wind speed and wind pressure.

[0050] Figure 6 The system generates a report for a specific project, displaying the label blocking and hard constraint boundary diagram. Detailed Implementation

[0051] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0052] This invention provides an automatic method for reviewing the physical consistency of parameters in hydrological and meteorological engineering reports, which involves a computer system performing the following data processing steps:

[0053] The process involves obtaining the hydrological and meteorological engineering report to be reviewed, and then using a document parsing engine to perform structured parsing on the main text, tables, figure captions, appendices, and calculation sheets in the report to obtain structured text content, numerical content, and parameter source location information. The document parsing engine can employ one or more of the following: the Docling document parsing framework, the Apache Tika document parsing component, the python-docx parsing component, or the LibreOffice conversion component, to convert engineering reports in .doc, .docx, and .pdf formats into structured text with chapters, tables, figure captions, and source location identifiers.

[0054] Based on the structured text content, a natural language processing model is used to identify engineering information, which includes at least the project type, design stage, site conditions, sea area location, elevation datum, and applicable standards. The natural language processing model can be a Qwen series large language model, a BERT-type named entity recognition model, a BiLSTM-CRF sequence labeling model, or a model based on a combination of rule dictionary and text classifier, and can be called in the local environment through LMStudio or OpenAI compatible interfaces.

[0055] Based on the project information, a parameter review requirement library is dynamically generated through a rule engine. The parameter review requirement library includes the hydrological and meteorological parameters that should be provided for the corresponding type of project, the corresponding applicable condition fields, and the source of the requirements. The rule engine can be implemented using Python-based regular expressions, keyword dictionaries, rule tables, and pattern matching programs, or it can be implemented using rule engine frameworks such as Drools and EasyRules. It is used to match the parameters to be reviewed and their triggering conditions according to the project type, design stage, and applicable specifications.

[0056] Hydrological and meteorological engineering parameters are extracted from the structured text and numerical content using an information extraction model to form a structured report parameter list. Each parameter item in the report parameter list includes at least the parameter type, parameter value, unit, source location, operating condition label, return period label, site label, benchmark label, specification version label, and verification status. The information extraction model can employ one or more of the following: Qwen series large language models combined with structured output constraints, BERT-CRF parameter entity extraction model, HanLP or spaCy entity recognition components, to extract wind speed, wind pressure, tide level, waves, current velocity, and their units, return periods, and source locations from the text, tables, and calculation sheets.

[0057] Based on a pre-defined physical constraint model library and parameter type mapping table, candidate physical relationship groups are constructed from the report parameter list according to the pre-defined physical relationships between parameters;

[0058] The candidate physical relationship group is subjected to multi-dimensional label matching judgment through a matching judgment matrix. The multi-dimensional labels include operating conditions, return period, site, benchmark, specification version, and statistical period.

[0059] When the candidate physical relationship group meets the multidimensional label matching conditions, the corresponding deterministic physical constraint model is called to perform calculation verification, obtain the physical calculation value and the deviation value, and save the calculation result as a locked state;

[0060] When the candidate physical relationship group does not meet the multidimensional label matching conditions, the physical formula calculation verification is not performed, a label mismatch prompt is generated, and the relevant parameter items are set to a state pending manual verification.

[0061] The verification status in the report parameter list is updated according to the deviation value and the preset threshold rule, and the parameter items that reach the preset error level are set to the unpassable status.

[0062] The physical calculation value, deviation value, source location, and verification status are output as review results to the review interface.

[0063] The following is a detailed implementation process of the present invention.

[0064] This invention provides an automatic method for verifying the physical consistency of parameters in hydrological and meteorological engineering reports, comprising the following steps:

[0065] First, the system acquires and parses the hydrological and meteorological engineering reports to be reviewed. The parsing includes main text paragraphs, tables, figure captions, appendices, calculation sheets, table of contents, chapter titles, and excerpts from the original text. During parsing, the system simultaneously retains page numbers, chapters, table numbers, row and column positions, and the original text context, ensuring that each parameter not only has a numerical value but also indicates its origin, applicable conditions, and related textual descriptions.

[0066] Second, identify the engineering information corresponding to the report. Engineering information should include at least the project name, project type, design stage, location, marine or land site conditions, elevation datum, statistical period, referenced standards and versions. This step determines the engineering context of the report, and then decides which parameters should be reviewed, which calculation relationships should be used, and which parameters can be compared within the same group.

[0067] Third, a parameter review requirement library is generated based on the project information. This library includes the parameter categories that should be provided for this type of project, the fields each parameter should have, applicable conditions, the source of the requirements, and missing item prompts. For example, marine photovoltaic, offshore wind power, and port or marine engineering projects typically require checking design wind parameters, wind pressure and conversion basis, tide level or design water level, design waves, design current velocity, applicable standards, and joint operating condition descriptions; while for land-based projects, requirements such as flood level, rainfall intensity, drainage conditions, or site elevation can be generated based on the project type.

[0068] Fourth, extract parameters from the report and create a report parameter list. Each parameter item should include at least the parameter name, parameter type, value, unit, source location, original text excerpt, working condition label, return period label, site label, elevation datum label, specification version label, statistical period, relationship group identifier, extraction confidence level, and verification status. Through this parameter list, information scattered across different chapters, tables, and appendices in the report is organized into calculable, traceable, and verifiable structured data.

[0069] Fifth, candidate physical relationship groups are established according to the preset physical constraint model. The system will not immediately calculate after discovering two values, but will first determine whether these parameters may belong to the same engineering relationship. For example, the wind speed-wind pressure relationship requires the same site, the same working condition, the same return period, and the same statistical period; the water level relationship requires the tidal level, water level, and elevation benchmark to be clearly defined; the wave relationship requires the simultaneous consideration of wave height, period, wave direction, water depth, return period, and sea state conditions.

[0070] Sixth, perform multi-dimensional label matching judgment on candidate physical relationship groups. The judgment content includes operating conditions, return period, site, benchmark, specification version, and statistical period. If necessary labels are missing, conflicting, or the source conditions cannot be confirmed, the system will not proceed with formula calculation, but will instead provide the reason for the mismatch and set the relevant parameters to a state pending manual review. This approach avoids directly comparing 50-year return period wind speed with 100-year return period wind pressure, water levels at different elevation benchmarks, or wave parameters under different operating conditions.

[0071] Seventh, when the tags match, the system uses fixed formulas or rules for calculation verification. Taking the wind speed-wind pressure relationship as an example, the system can use W=V... 2 / K calculates the wind pressure, where K is determined by the unit system, air density, statistical period, specification version, or project configuration. After the calculation is completed, the system will report a comparison between the given value and the calculated value, obtain the percentage deviation, and update the parameter status according to the threshold rules to verified, warning, calculation error, or unacceptable.

[0072] Eighth, the calculation and verification results should be saved as a fixed basis that the review generation module must use. These results should include at least the alarm type, severity level, parameter identifier, source location, report setpoint, calculated value, percentage deviation, calculation model, explanation, recommendations, and status identifier. The review generation module may organize natural language comments based on this, but it must not recalculate, replace, delete, or downgrade the already determined numerical conclusions.

[0073] Ninth, implement process control for parameters that reach the error level. Parameters set to "unacceptable" must not be included in the report review approval list, structured results library, historical case library, or subsequent simulation calculation modules. This status can only be restored after the report is revised and re-approved, or after manual review confirming the application of special rules. Therefore, the review results not only serve as a warning but also affect subsequent data usage.

[0074] Example: Review of a project's hydrological and meteorological report

[0075] This embodiment uses a "Hydrological and Meteorological Report of a Marine Photovoltaic Project" as the report to be reviewed. This report is a hydrological and meteorological report for a marine photovoltaic project. After the system activates the standard-driven parameter review mode, it first parses the report text, tables, figure captions, and appendices, extracting the project name, project type, design stage, marine location, site conditions, elevation datum, referenced standards, and hydrological and meteorological parameters.

[0076] like Figure 1 As shown, the system's review process includes steps such as report parsing, engineering information identification, generation of a parameter review requirement library, generation of a report parameter list, construction of candidate relationship groups, tag matching judgment, fixed formula calculation verification, status update and process control, and output of review results. In this embodiment, these steps revolve around the wind, tide, wave, current, and regulatory basis of a specific project.

[0077] like Figure 2 and Figure 3 As shown, the system identified the project as a marine photovoltaic project and generated a patent examination file. This file displays the scope of the additional examination completed this time, including the parameters that needed to be examined, the parameters already found in the report, the deterministic relationship verification performed in this actual execution, and the results of the fixed basis. The system clearly distinguishes between ordinary examination opinions and the examination results of normatively driven parameters in this file.

[0078] like Figure 4As shown, the system generates a parameter review requirement library based on the engineering information of a project. For marine photovoltaic projects, the requirement library includes design wind parameters, wind pressure and conversion basis, applicable standards and versions, tide level or design water level, design waves, design current velocity, and a description of combined operating conditions. Each type of parameter includes the fields to be provided, the source of the requirement, and the review status. For example, design wind parameters should include wind speed, return period, statistical period, data source or calculation basis; wind pressure and conversion basis should include wind pressure, corresponding wind speed, conversion relationship or standard basis.

[0079] The system extracts parameters from a project report and generates a list of report parameters. The list records the design wind speed and design wind pressure for 25-year and 50-year return periods, along with their source segments, return periods, relationship group identifiers, and verification status. For the 50-year return period wind speed-pressure relationship, the system identifies the reported wind speed as 38.0 m / s and the reported wind pressure as 0.90 kN / m². 2 Since both have the same site, the same design conditions and the same return period, they meet the label matching conditions of the wind speed-wind pressure relationship group.

[0080] like Figure 5 As shown, the system performs a fixed formula calculation and verification using W=V² / 1600, yielding a calculated 50-year return period wind pressure value of 38.0. 2 / 1600=0.9025kN / m 2 The report gives a value of 0.90 kN / m. 2 Compared to the calculated value of 0.9025 kN / m 2 The deviation is approximately 0.28%, which is below the passing threshold. Therefore, the system sets the status of this relationship group to verified. This calculated value, deviation value, source location, and verification status are written into the review file as fixed criteria and cannot be modified by the review generation module.

[0081] If the same report or a copy of the report lists a 50-year return period wind speed of 38.0 m / s but the corresponding wind pressure is written as 0.80 kN / m², then... 2 If a significant deviation is found within the same relation group, the system will mark the wind pressure parameter as either an abnormal calculation or unacceptable. Unacceptable parameters must not be included in the approved list, structured results library, or subsequent simulation calculation modules. If the report only contains 50-year return period wind speed and 100-year return period wind pressure, the system will stop calculation due to a mismatch in return period labels and output a message indicating that manual review is required.

[0082] This embodiment illustrates that the present invention does not simply calculate a single formula in the report, but rather comprehensively reviews the physical consistency of hydrological and meteorological report parameters through engineering information, parameter requirements, parameter lists, relationship groups, label matching, verification of fixed formula calculations, and state control. This technical solution can improve the ability to detect hidden parameter errors in actual engineering report review and reduce the instability risks caused by the direct involvement of the generated model in numerical judgment.

[0083] This invention also provides an automatic review system for the physical consistency of parameters in hydrological and meteorological engineering reports, which performs the method described above, including:

[0084] The document parsing module is used to obtain the hydrological and meteorological engineering report to be reviewed and parse the main text, tables, figure captions, appendices and calculation sheets to obtain the text content, numerical content and parameter source location information;

[0085] The engineering information identification module is used to identify engineering information based on the parsed text content through a natural language processing model. The engineering information includes at least the engineering type, design stage, site conditions, sea area location, elevation datum, and applicable specifications.

[0086] The parameter review requirement library generation module is used to dynamically generate a parameter review requirement library based on the project information through a rule engine. The parameter review requirement library includes the hydrological and meteorological parameters that should be provided for the corresponding type of project, the corresponding applicable condition fields, and the source of the requirements.

[0087] The report parameter list generation module is used to extract hydrological and meteorological engineering parameters from text and numerical content through an information extraction model and form a structured report parameter list. Each parameter item in the report parameter list includes at least parameter type, parameter value, unit, source location, working condition label, return period label, site label, benchmark label, specification version label, and verification status.

[0088] The candidate physical relationship group construction module is used to construct candidate physical relationship groups from the report parameter list based on the preset physical constraint model library and parameter type mapping table, according to the preset physical relationships between parameters;

[0089] The multidimensional label matching and judgment module is used to perform multidimensional label matching and judgment on candidate physical relationship groups through a matching and judgment matrix. The multidimensional labels include operating conditions, recurrence period, site, benchmark, specification version and statistical period.

[0090] The fixed formula calculation and verification module is used to call the deterministic physical constraint model to perform calculation and verification under the condition that the label matching is satisfied, to obtain the physical calculation value and the deviation value, and to save the calculation result as a locked state. The calculation result after being locked is prohibited from being rewritten by the subsequent generated model.

[0091] The status update and process control module is used to update the verification status according to the deviation value and preset threshold rules, and set the parameter items that reach the preset error level to the unpassable status. The parameter items in the unpassable status are prohibited from entering the report review approval list, the structured results library and the subsequent automatic calculation module.

[0092] The review result generation module is used to output the physical calculation value, deviation value, source location and verification status as review results to the review interface.

[0093] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0094] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. An automatic method for verifying the physical consistency of parameters in hydrological and meteorological engineering reports, characterized in that, The computer system performs the following data processing steps: Obtain the hydrological and meteorological engineering report to be reviewed, and use a document parsing engine to perform structured parsing on the main text, tables, figure captions, appendices and calculation sheets in the report to obtain structured text content, numerical content and parameter source location information; Based on the structured text content, a natural language processing model is used to identify engineering information, which includes at least the project type, design stage, site conditions, sea area location, elevation datum, and applicable standards. Based on the project information, a parameter review requirement library is dynamically generated through a rule engine. The parameter review requirement library includes the hydrological and meteorological parameters that should be provided for the corresponding type of project, the corresponding applicable condition fields, and the source of the requirements. Hydrological and meteorological engineering parameters are extracted from the structured text content and numerical content using an information extraction model to form a structured report parameter list. Each parameter item in the report parameter list includes at least the parameter type, parameter value, unit, source location, operating condition label, return period label, site label, benchmark label, specification version label, and verification status. Based on a pre-defined physical constraint model library and parameter type mapping table, candidate physical relationship groups are constructed from the report parameter list according to the pre-defined physical relationships between parameters; The candidate physical relationship group is subjected to multi-dimensional label matching judgment through a matching judgment matrix. The multi-dimensional labels include operating conditions, return period, site, benchmark, specification version, and statistical period. When the candidate physical relationship group meets the multidimensional label matching conditions, the corresponding deterministic physical constraint model is called to perform calculation verification, obtain the physical calculation value and the deviation value, and save the calculation result as a locked state; When the candidate physical relationship group does not meet the multidimensional label matching conditions, the physical formula calculation verification is not performed, a label mismatch prompt is generated, and the relevant parameter items are set to a state pending manual verification. The verification status in the report parameter list is updated according to the deviation value and the preset threshold rule, and the parameter items that reach the preset error level are set to the unpassable status. The physical calculation value, deviation value, source location, and verification status are output as review results to the review interface.

2. The automatic verification method for physical consistency of parameters in hydrological and meteorological engineering reports according to claim 1, characterized in that, The process of using a natural language processing model to identify engineering information includes: extracting information such as project type, design stage, province, marine or land-based site conditions, marine location, elevation datum, statistical period, and standard number and name cited in the report from the report name, project overview, technical basis section, design parameter section, table title, figure caption, appendix name, and parameter context using named entity recognition and text classification models. The identification results are then used as input to generate a parameter review requirement library and matching judgment rules.

3. The automatic verification method for physical consistency of parameters in hydrological and meteorological engineering reports according to claim 1, characterized in that, The parameter review requirement library is dynamically generated based on engineering information. For marine photovoltaic, offshore wind power, port or marine engineering projects, the parameter review requirement library shall include at least one or more of the following: design wind parameters, wind pressure and conversion basis, tide level or design water level, design waves, design current velocity, applicable specifications and versions, and joint working condition descriptions. It shall also configure the required fields, applicable conditions and missing item prompts for each parameter.

4. The automatic verification method for physical consistency of parameters in hydrological and meteorological engineering reports according to claim 1, characterized in that, The parameter items in the report parameter list also include original text excerpts, chapters or page numbers, table rows and columns, statistical periods, extraction confidence levels, and relationship group identifiers. The relationship group identifiers are used to group different physical quantities from the same site, the same working condition, the same return period, the same elevation datum, or the same original text context into the same candidate physical relationship group.

5. The automatic verification method for physical consistency of parameters in hydrological and meteorological engineering reports according to claim 1, characterized in that, The multidimensional label matching judgment is achieved through a matching judgment matrix, rule base, key-value mapping or database table; when any necessary label in the working condition, recurrence period, site, benchmark, specification version or statistical period is missing or conflicting, the physical formula calculation verification is not performed, a label mismatch prompt is generated and the relevant parameter items are set to a state pending manual verification.

6. The automatic verification method for physical consistency of parameters in hydrological and meteorological engineering reports according to claim 1, characterized in that, The deterministic physical constraint model includes a wind speed-wind pressure physical constraint model, which is based on W=V. 2 / K is used to calculate wind pressure, where W is wind pressure, V is wind speed, and K is a conversion factor related to the unit system, air density, statistical period, or standard version.

7. The automatic verification method for physical consistency of parameters in hydrological and meteorological engineering reports according to claim 1, characterized in that, The deterministic physical constraint model also includes one or more of the following: wave parameter physical constraint model, water level condition physical constraint model, elevation datum conversion constraint model, tide level applicable condition constraint model, and cross-chapter consistency constraint model for parameters of the same origin.

8. The automatic verification method for physical consistency of parameters in hydrological and meteorological engineering reports according to claim 1, characterized in that, The verification status includes pending verification, verified, prompt, calculation error, missing item or pending confirmation, pending manual review, and unpassable. When a parameter item is set to an unpassable status, the system controls the corresponding parameter item to not enter the report review approval list, structured results library, historical case positive example library, or subsequent automatic calculation module.

9. An automatic system for reviewing the physical consistency of parameters in hydrological and meteorological engineering reports, comprising the method described in any one of claims 1 to 8, characterized in that, include: The document parsing module is used to obtain the hydrological and meteorological engineering report to be reviewed and parse the main text, tables, figure captions, appendices and calculation sheets to obtain the text content, numerical content and parameter source location information; The engineering information identification module is used to identify engineering information based on the parsed text content through a natural language processing model. The engineering information includes at least the engineering type, design stage, site conditions, sea area location, elevation datum, and applicable specifications. The parameter review requirement library generation module is used to dynamically generate a parameter review requirement library based on the project information through a rule engine. The parameter review requirement library includes the hydrological and meteorological parameters that should be provided for the corresponding type of project, the corresponding applicable condition fields, and the source of the requirements. The report parameter list generation module is used to extract hydrological and meteorological engineering parameters from text and numerical content through an information extraction model and form a structured report parameter list. Each parameter item in the report parameter list includes at least parameter type, parameter value, unit, source location, working condition label, return period label, site label, benchmark label, specification version label, and verification status. The candidate physical relationship group construction module is used to construct candidate physical relationship groups from the report parameter list based on the preset physical constraint model library and parameter type mapping table, according to the preset physical relationships between parameters; The multidimensional label matching and judgment module is used to perform multidimensional label matching and judgment on candidate physical relationship groups through a matching and judgment matrix. The multidimensional labels include operating conditions, recurrence period, site, benchmark, specification version and statistical period. The fixed formula calculation and verification module is used to call the deterministic physical constraint model to perform calculation and verification under the condition that the label matching is satisfied, to obtain the physical calculation value and the deviation value, and to save the calculation result as a locked state. The calculation result after being locked is prohibited from being rewritten by the subsequent generated model. The status update and process control module is used to update the verification status according to the deviation value and preset threshold rules, and set the parameter items that reach the preset error level to the unpassable status. The parameter items in the unpassable status are prohibited from entering the report review approval list, the structured results library and the subsequent automatic calculation module. The review result generation module is used to output the physical calculation value, deviation value, source location and verification status as review results to the review interface.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of any one of claims 1 to 8.

Images