Rule engine-based automatic auditing method and system for water conservancy model

Abstract

This invention relates to a method and system for automated review of water conservancy models based on a rule engine, belonging to the field of model application and automated review technology. The invention constructs a basic review rule library for water conservancy models containing general rules and generates a complete set of dedicated review rules through rule instantiation; it utilizes a large water conservancy model to perform multi-format semantic parsing of the original submitted materials, and automatically generates standardized model operation strategies based on the large model parsing results and verification indicators in the dedicated review rule set, driving the automated trial operation of the water conservancy model; the rule engine collects real-time data on the trial operation status and verification indicators of the water conservancy model, calls the dedicated review rule set bound to the model, and executes rule judgments in parallel by review dimension; if the review is successful, a standardized review report is generated; if the review fails, anomalies are automatically marked and a manual review process is triggered. This invention significantly improves the efficiency and accuracy of water conservancy model review.

Description

Technical Field

[0001] This invention relates to an automated review method and system for water conservancy models based on a rule engine, belonging to the field of model application and automated review technology. Background Technology

[0002] The large-scale water conservancy model is derived from a general large-scale model by fine-tuning and training it by integrating national standards, industry norms, and historical audit data for water conservancy projects. It has the ability to understand water conservancy professional terminology (such as "peak flow" and "roughness coefficient"), parse materials in multiple formats (GeoJSON, NetCDF, tiff, Excel, Word, PDF), and reason about model operation strategies.

[0003] The existing hydraulic model review process has the following technical problems, making it unable to meet the review needs of the massive number of models in the hydraulic industry:

[0004] 1. Efficiency bottleneck of manual review: Reviewers need to check model materials item by item, manually configure operating parameters, and manually perform model verification. The review of a single model takes several hours to several days, making it difficult to handle massive review tasks. In addition, the lack of a unified rule system leads to subjective differences in judgment among different reviewers on the same technical indicators, resulting in poor standardization and repeatability of review conclusions.

[0005] 2. Poor adaptability of general rules: The preset general basic review rules cannot be directly adapted to different types of water conservancy models (flood control, hydrology, hydraulics, etc.), and the rule conditions and parameter thresholds need to be manually adjusted one by one, resulting in low rule reusability;

[0006] 3. Existing methods only review static materials and lack dynamic verification of the model's operational effectiveness: Existing review methods based on rule engines or large models + rule engines review static materials (parameter files, documents, images, etc.) uploaded by the submitter, and determine whether the materials are compliant through rule matching; however, one of the core requirements of water conservancy model review is to verify whether the model can be executed correctly under real operating conditions and whether the results comply with water conservancy engineering specifications—this dimension of "operational effectiveness" review cannot be achieved by existing technical solutions and still requires manual operation of the model and item-by-item verification;

[0007] 4. Existing large-scale model + rule engine fusion solutions employ a parallel architecture, with the results of the two being merged independently. This lacks a serial coupling mechanism where large-scale model parsing drives rule instantiation: a rule engine alone cannot parse unstructured submissions such as Word and PDF documents; a large-scale model alone lacks the precise structured execution capability for review logic, and the two do not form a collaborative technical architecture. Current solutions combining large-scale models and rule engines typically adopt a parallel architecture where "the large-scale model and rule engine process input materials independently, and then merge their outputs." The rules in the rule engine remain manually preset static rules.

[0008] 5. Lack of automated solutions for model trial runs and no closed-loop review process: Model trial runs require manual configuration of the operating environment, input parameters, and execution steps; there is no standardized anomaly localization mechanism after a failure, leaving submitters without clear direction for modifications, leading to repeated re-reviews. Failed reviews are simply rejected without precise marking of the anomaly location and cause, lacking a standardized "anomaly marking - manual review - result feedback" closed-loop mechanism. Summary of the Invention

[0009] The purpose of this invention is to overcome the above-mentioned shortcomings and provide an automated review method and system for water conservancy models based on a rule engine. Designed to meet the review needs of massive models in the water conservancy industry, it integrates the semantic parsing capabilities of Large Language Model (LLM) with the logical execution capabilities of a rule engine (RE) to solve the problems of low efficiency and inconsistent standards in traditional manual review.

[0010] The technical solution adopted in this invention is as follows:

[0011] The automated review method for water conservancy models based on a rule engine includes the following steps:

[0012] S1. Construct a basic review rule library for water conservancy models that includes general rules;

[0013] S2. Rule Instantiation: Receives user-uploaded water conservancy model submission materials, automatically identifies the material type and model application scenario, drives the rule engine to retrieve suitable rules from the rule library, performs fine-grained rule instantiation, and generates a complete set of fine-grained exclusive review rules bound to the model;

[0014] S3. Utilize the large-scale water conservancy model to perform multi-format semantic parsing on the original submitted materials, extract the core parameters, operational boundary conditions, and key performance indicators of the model in a structured manner, and output the parsing results in a standardized format as the driving input for triggering the instantiation of the rule engine;

[0015] S4. Based on the large model analysis results and the verification indicators in the exclusive audit rule set, automatically generate standardized model operation strategies and drive the automated trial operation of the water conservancy model;

[0016] S5. The rule engine collects real-time data on the trial operation status and verification indicators of the water conservancy model, calls the exclusive audit rule set bound to the model, and executes the rule judgment in parallel according to the audit dimension. If the audit is successful, a standardized audit report is generated; if the audit fails, anomalies are automatically marked and a manual review process is triggered.

[0017] S6. After manual review, the review will be revised and resubmitted according to the judgment results, or the model will be modified and resubmitted, or the case will be directly rejected, forming a complete closed loop of "automatic review - anomaly marking - manual review - result feedback".

[0018] The method also includes batch review scheduling, and a batch automated review process for water conservancy models; specifically, it includes: task queue construction, concurrent scheduling, breakpoint resume review, and batch result aggregation.

[0019] In the above method, the rules in step S1 are stored in JSON structured format and managed according to five dimensions: material completeness, format compliance, parameter rationality, operational effectiveness, and result standardization; the rules are indexed in three dimensions according to the review dimension, material type, and model type.

[0020] The fine-grained instantiation of rules in step S2 includes condition refinement, parameter assignment, and logical combination. In step S2, the logic consistency of the instantiated exclusive audit rule set needs to be checked: whether there are contradictions between rules, whether there are duplicate rules, and whether the CRITICAL (major defect) level rules cover all five audit dimensions. After the check is passed, the exclusive audit rule set is bound and stored with the model submission ID.

[0021] The standardized model operation strategy in step S4 includes parameter configuration, execution steps, verification metrics, and termination conditions.

[0022] The rule-based water conservancy model automated review system comprises a data support layer, a user interaction layer, a core engine layer, and a business execution layer.

[0023] The data support layer is used for storing the basic review rule base for water conservancy models, the water conservancy professional knowledge base, the model material library, and the review database.

[0024] The user interaction layer includes a material submission and upload module, a review progress and review report viewing module, and a manual review processing module.

[0025] The core engine layer includes a large water conservancy model engine and a rules engine. The rules engine automatically identifies the material type and model application scenario based on the materials submitted by the user's uploaded water conservancy model, drives the rules engine to retrieve the appropriate rules from the rule library, performs fine-grained instantiation of the rules, and generates a complete set of fine-grained exclusive review rules bound to the model.

[0026] The business execution layer includes a material analysis module, a strategy generation module, a trial operation module, an audit and judgment module, an anomaly marking module, and a review management module. The material analysis module uses a large-scale water conservancy model to perform multi-format semantic analysis on the original submitted materials, extracting core model parameters, operational boundary conditions, and key performance indicators in a structured manner. The analysis results are output in a standardized format as the driving input for triggering the rule engine instantiation. The strategy generation module automatically generates standardized model operation strategies based on the large-scale model analysis results and verification indicators in a dedicated audit rule set. The trial operation module is responsible for the automated trial operation of the water conservancy model. The audit and judgment module collects real-time trial operation status data and verification indicator data from the rule engine, calls the dedicated audit rule set bound to the model, and executes rule judgments in parallel according to audit dimensions. If the audit is successful, a standardized audit report is generated. The anomaly marking module automatically marks anomalies for failed audits and triggers a manual review process. The review management module is responsible for manual review. After manual review, based on the judgment results, it executes rule correction and re-review, model modification and re-review, or direct rejection, forming a complete "automatic review - anomaly marking - manual review - result feedback" audit closed loop.

[0027] The aforementioned rule engine-based automated review system for water conservancy models also includes a batch review scheduling module, which performs batch automated review steps for water conservancy models, completes task queue construction, concurrent scheduling, breakpoint resume review, and batch result aggregation.

[0028] The beneficial effects of this invention are:

[0029] (1) Solve the problem of poor rule adaptability: Through dynamic instantiation of the rule engine, 27 general basic rules can be automatically expanded into hundreds of fine-grained exclusive rules for specific model types and scenarios, without the need for manual adjustment of each set, and the reusability of rules is greatly improved;

[0030] (2) Achieve deep collaboration between the big model and the rule engine: The big model is responsible for the semantic parsing of unstructured / semi-structured materials and the structured extraction of key information, while the rule engine is responsible for accurate logical judgment and execution. The two complement each other and together constitute a complete automated review technology architecture, overcoming the inherent defects of using either technology alone.

[0031] (3) To realize the automatic review of the effectiveness of the operation of the water conservancy model: In addition to the review of static material rules, an automated trial operation is introduced. The system automatically generates operation strategies, drives the actual operation of the model, collects operation data and makes rule judgments, extending the review capability to the core dimension of "whether the model can run correctly" which cannot be covered by existing technical solutions, thus eliminating the efficiency bottleneck of manual model verification.

[0032] (4) Achieve full-process automation: From material format recognition, rule instantiation, large model parsing, strategy generation, automated model trial operation, to review judgment and report generation, no manual intervention is required throughout the entire process. Only when the review fails is a minimal manual review intervention triggered, freeing reviewers from repetitive operations.

[0033] (5) Establish a unified review rule system: The basic rule base and dynamic instantiation mechanism together ensure that all models are reviewed according to a unified standard, eliminating the subjective bias of manual review, and the review conclusions are repeatable and traceable;

[0034] (6) Construct a precise anomaly marking and closed-loop review mechanism: After the review fails, a precise marking list containing anomaly type, anomaly location and cause analysis will be automatically output to provide clear direction for the submitter to make modifications. At the same time, it supports three-path processing of manual review (rule correction review / submitter modification review / rejection) to form a complete review closed loop;

[0035] (7) Supports massive batch review: Batch task queuing scheduling + server-side concurrent execution + breakpoint resume review mechanism, ensuring that the system can stably handle the review business needs of massive water conservancy models, and at the same time supports real-time monitoring of batch review progress;

[0036] (8) Wide range of material format compatibility: It is compatible with commonly used spatial data formats in water conservancy such as GeoJSON, NetCDF, and TIFF, as well as general formats such as Excel, CSV, Word, and PDF, to meet the material submission and review requirements of water conservancy models of different types and application scenarios. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the dynamic instantiation process of the rule engine in this invention.

[0038] Figure 2 This is a flowchart illustrating the automated review and closed-loop verification process of this invention.

[0039] Figure 3 This is a flowchart of the batch review task scheduling process of the present invention;

[0040] Figure 4 This is a diagram of the overall system architecture of the present invention. Detailed Implementation

[0041] The following description, in conjunction with specific embodiments, provides further details.

[0042] Example 1: Automated Review System for Hydraulic Models Based on Rule Engine (e.g., ...) Figure 4 It includes a data support layer, a user interaction layer, a core engine layer, and a business execution layer.

[0043] The system is based on a B / S architecture, with the server side handling core calculations and rule execution, and the browser side providing the user interaction interface.

[0044] The data support layer is used for storing the basic review rule base for water conservancy models, the water conservancy professional knowledge base, the model material library, and the review database.

[0045] The user interaction layer includes a material submission and upload module, a review progress and review report viewing module, and a manual review processing module.

[0046] The core engine layer includes a large water conservancy model engine and a rules engine. The rules engine automatically identifies the material type and model application scenario based on the materials submitted by the user's uploaded water conservancy model, drives the rules engine to retrieve the appropriate rules from the rule library, performs fine-grained instantiation of the rules, and generates a complete set of fine-grained exclusive review rules bound to the model.

[0047] The business execution layer includes a material analysis module, a strategy generation module, a trial operation module, an audit and judgment module, an anomaly marking module, and a review management module. The material analysis module uses a large-scale water conservancy model to perform multi-format semantic analysis on the original submitted materials, extracting core model parameters, operational boundary conditions, and key performance indicators in a structured manner. The analysis results are output in a standardized format as the driving input for triggering the rule engine instantiation. The strategy generation module automatically generates standardized model operation strategies based on the large-scale model analysis results and verification indicators in a dedicated audit rule set. The trial operation module is responsible for the automated trial operation of the water conservancy model. The audit and judgment module collects real-time trial operation status data and verification indicator data from the rule engine, calls the dedicated audit rule set bound to the model, and executes rule judgments in parallel according to audit dimensions. If the audit is successful, a standardized audit report is generated. The anomaly marking module automatically marks anomalies for failed audits and triggers a manual review process. The review management module is responsible for manual review. After manual review, based on the judgment results, it executes rule correction and re-review, model modification and re-review, or direct rejection, forming a complete "automatic review - anomaly marking - manual review - result feedback" audit closed loop.

[0048] The aforementioned rule engine-based automated review system for water conservancy models also includes a batch review scheduling module, which performs batch automated review steps for water conservancy models, completes task queue construction, concurrent scheduling, breakpoint resume review, and batch result aggregation.

[0049] Example 2: An automated review method for water conservancy models based on a rule engine, comprising the following steps:

[0050] S1. Construct a basic audit rule base for water conservancy models containing general rules:

[0051] (1) Rule design and storage structure definition

[0052] Based on five audit dimensions—material completeness, format compliance, parameter rationality, operational effectiveness, and deliverable standardization—27 general basic audit rules were formulated. Each rule is stored in JSON format and includes fields such as a unique rule identifier, audit dimension, applicable model type, applicable material type, judgment condition template, judgment criteria, and rule severity level.

[0053] The core design of the rule storage structure lies in the separation of condition_template and params_placeholder: the basic rule only stores the "structural skeleton" of the condition (including placeholders), while the specific numerical parameters are automatically filled in by the system from the parameter threshold database according to the model type and scenario during instantiation. This design allows the same basic rule to derive multiple fine-grained rules for different model types and scenarios, realizing the automatic expansion of "one basic rule → multiple exclusive rules", which is the technical foundation of the dynamic instantiation capability of this method.

[0054] (2) Rule-based classification storage and index creation

[0055] The 27 basic rules are indexed in three dimensions according to the review dimensions, material types, and model types. The rule engine supports quick retrieval of appropriate rule subsets by single or multi-dimensional combination.

[0056] (3) Pre-set instantiation interface

[0057] The rule engine reserves an instantiation interface for each basic rule, supporting external input of instantiation parameters (material_type, model_type, scenario_params, etc.). After the interface is called, it triggers condition refinement, parameter assignment, and logic combination to generate fine-grained instantiation rules.

[0058] S2. Rule instantiation (e.g.) Figure 1 ): Receives user-uploaded materials for water conservancy models, automatically identifies the material type and model application scenario, drives the rule engine to retrieve suitable rules from the rule library, performs fine-grained rule instantiation, and generates a complete set of fine-grained, exclusive review rules bound to the model.

[0059] (1) Identification of Submitted Material Types

[0060] Users upload hydraulic models and submit materials via their browsers (batch uploads are supported). The server-side material recognition module automatically performs the following recognition tasks:

[0061] a. File format identification: Identify the format type based on the file extension and header bytes (e.g., .nc → NetCDF, .geojson → GeoJSON, .xlsx → parameter table, .pdf → report, etc.).

[0062] b. Material type classification: The identified files are categorized into parameter files, boundary condition files, mesh files, operation logs, result reports, etc.

[0063] c. Model application scenario identification: Analyze key fields in the materials (such as model type identifier field, project name, etc.), combine with file naming conventions, determine the model application scenario (flood control model / hydrological model / hydraulic model / dispatch model, etc.), encapsulate the identification results in JSON format, and transmit them to the rule engine.

[0064] (2) Retrieval of basic rules

[0065] The rule engine receives the material identification results and uses the list of model_type (model type) and material_type (material type) as query conditions to retrieve a subset of suitable basic rules from the rule base to form a set of rules to be instantiated (usually 15 to 20 basic rules).

[0066] (3) Dynamic instantiation of rules

[0067] The rule engine iterates through each basic rule in the rule set to be instantiated, performing the following three operations:

[0068] a. Condition Refinement: Based on the material type and model application scenario, the general condition descriptions in the rule's condition_template are refined into specific field-level judgment conditions. For example, the basic rule "parameter file field integrity" is refined for flood control models into non-empty verification conditions for specific fields such as "upstream inflow," "riverbed roughness," and "grid resolution."

[0069] b. Parameter assignment: From the system's built-in hydraulic model parameter threshold database, query the reasonable range of each parameter according to the model type and scenario, and fill the params_placeholder in the basic rules (such as replacing min_val (minimum value) and max_val (maximum value) with specific values).

[0070] c. Logical combination: When multiple constraints are involved in the same field (such as satisfying both range constraints and format constraints at the same time), multiple sub-rules are combined into a compound rule using AND-OR logic.

[0071] Once a single water conservancy model is instantiated, its dedicated review rule set typically contains 200 to 500 fine-grained rules (depending on the model type and the number of submitted materials).

[0072] Specific example of rule instantiation: Basic rule R-012 (boundary condition integrity check), which generates three fine-grained rules after instantiation of the flood control model:

[0073] Table 1

[0074] .

[0075] Different fine-grained rules are generated after the hydrological model is instantiated, which reflects the core value of dynamic instantiation.

[0076] In existing technologies, "dynamic rules" typically refer to pre-classifying and storing rules by level (e.g., basic level / dynamic level / industry level), and calling different levels of rules as needed. The rules themselves are already defined and complete when stored, and no modification is needed when calling them. The "dynamic instantiation" in this invention's method refers to the fact that the basic rules stored in the rule base are incomplete (including parameter placeholders). During each review, the system performs real-time parameter filling and logical combination of the basic rules based on the characteristics of the currently submitted materials (model type, material type, scene parameters), generating a unique and complete rule specific to this review task—the rule content is dynamically generated at runtime, rather than statically stored. This is the core technical meaning of the "dynamic rule instantiation" in this method.

[0077] (4) Rule set logic verification

[0078] The system performs a logical consistency check on the instantiated dedicated audit rule set: checking for contradictions between rules (such as logical conflicts between upper and lower limits of the same field), duplicate rules (rule deduplication), and whether CRITICAL (critical defect) level rules cover all five audit dimensions. If the check passes, the dedicated audit rule set is bound and stored with the model submission ID; if the check fails, the conflicting rule information is fed back to the rule engine, triggering a re-instantiation of the conflicting rules (maximum of 3 retries; if all 3 retries fail, an alert is triggered and the case is transferred to manual processing).

[0079] S3. Utilize the large-scale water conservancy model to perform multi-format semantic parsing on the original submitted materials, extract the core parameters, operational boundary conditions, and key performance indicators of the model in a structured manner, and output the parsing results in a standardized format as the driving input for triggering the instantiation of the rule engine.

[0080] The large model receives original submissions of hydraulic model data uploaded by users and employs corresponding parsing strategies for different file formats.

[0081] a. Structured files (Excel parameter tables, CSV data files): Directly read table data, and the large model identifies field semantics based on the water conservancy professional knowledge base (handling cases of non-standard field names, mixed use of Chinese and English, etc.) to complete field value extraction;

[0082] b. Semi-structured files (NetCDF, GeoJSON, TIFF spatial data): The corresponding format parsing library is called to extract metadata and numerical data. The large model performs semantic annotation on the extracted results to identify the corresponding hydraulic parameter types (such as watershed boundaries, DEM elevation, water level time series, etc.).

[0083] c. Unstructured files (Word documentation, PDF results report): The large model is driven by SystemPrompt, a dedicated tool for water conservancy review, to perform semantic parsing and extract information such as key technical parameters, engineering feature descriptions, and result index values.

[0084] The large model completes three core extraction tasks:

[0085] Extraction of core model parameters: such as hydrological station codes, model grid resolution, riverbed roughness coefficient, etc.

[0086] Model operation boundary condition extraction: such as inflow boundary flow process, outflow boundary water level condition, initial water level condition, simulation period, etc.;

[0087] Key indicators extracted from model results: such as peak flow, peak water level, maximum inundated area, and water conservation error.

[0088] After the large model is parsed, the extracted results are output in a standardized JSON format (see the JSON illustration in the basic knowledge section). This JSON output serves as the input parameter for subsequent steps two (strategy generation) and rule engine instantiation, and is the standardized interface for information transmission between the "large model → rule engine" in the serial driving chain of this invention. This is fundamentally different from the use of "large model output as the final judgment result" in existing parallel architecture schemes: in this invention, the large model output is intermediate structured data that drives downstream rule instantiation and strategy generation, rather than the final review conclusion. All extracted results are standardized and encapsulated into structured JSON.

[0089] S4. Based on the large model analysis results and the verification indicators in the exclusive audit rule set, automatically generate standardized model operation strategies and drive the automated trial operation of the water conservancy model;

[0090] Based on the standardized parsed JSON output from the large model, and simultaneously reading the CRITICAL (Major Flaw) level rules from the dedicated audit rule set. The field automatically generates a standardized model execution strategy (JSON format), which consists of four parts:

[0091] a. Parameter configuration: Map the core_params and boundary_conditions in the parsing results to the parameter format required by the model running environment, and automatically complete the parameter assignment;

[0092] b. Execution step sequence: In accordance with the hydraulic model operation specifications, a standardized list of model execution steps is generated, with each step including an action description, input dependencies, and timeout limit;

[0093] c. Verification indicator list: Extract the indicators that need to be monitored in real time from the dedicated audit rule set, generate an indicator list with monitoring thresholds, and connect it with the trial operation module;

[0094] d. Termination conditions: Define normal termination conditions and forced termination conditions.

[0095] After the strategy is generated, the system performs parameter integrity verification (checking whether all required parameters have been assigned values) and indicator coverage verification (checking whether all indicators required by the audit rules have corresponding monitoring items in the strategy). After the verification is passed, the model is driven to enter automated trial operation.

[0096] This step is the core and unique aspect of this invention, distinguishing it from all existing similar review methods. Existing rule engine review, large model review, and large model + rule engine integrated review solutions all review static materials uploaded by the submitter, determining compliance through rule matching or semantic understanding; no existing solution actually drives the reviewed model to run and reviews the results during the review process. This step, through automated trial operation, extends the review capability from "whether the material description is compliant" to "whether the model's actual operation is effective," achieving fully automated review of the effectiveness dimension of water conservancy model operation.

[0097] The trial operation module executes the automated trial operation of the hydraulic model step by step according to the generated operation strategy, collecting two types of data in real time during the process:

[0098] Runtime status data: The start time, completion time, whether an error occurred (and the error code), and whether a timeout occurred for each execution step are summarized into a runtime status time sequence record, and the sampling frequency is once at the end of each step.

[0099] Verify indicator data: According to the list of verification indicators in the strategy, collect the real-time calculated values ​​of each indicator at the specified sampling time to form indicator time series data.

[0100] All collected data is written to the audit database in real time and pushed synchronously to the rule engine audit execution module. At the same time, the real-time progress of the trial run (current execution steps, percentage completed, and real-time trends of key indicators) is displayed on the browser.

[0101] S5. The rule engine collects real-time data on the trial operation status and verification indicators of the water conservancy model, calls the exclusive audit rule set bound to the model, and executes the rule judgment in parallel according to the audit dimension. If the audit is successful, a standardized audit report is generated; if the audit fails, anomalies are automatically marked and a manual review process is triggered.

[0102] The rules engine receives status data and verification indicator data collected during the trial run, calls the exclusive audit rule set bound to the model, and executes rule judgments in parallel by grouping according to audit dimensions.

[0103] Approval criteria: If the model runs normally throughout (no errors, no timeouts), and all validation metric values ​​fall within the judgment range of the corresponding rules in the dedicated rule set, then the approval is deemed successful. The system automatically generates a standardized audit report, which includes: a summary of rule execution (number of approved rules / total number of rules), audit conclusions for each dimension, a comparison table of validation metrics (actual values ​​vs. standard ranges), audit conclusions and recommendations. The audit report is synchronized to the browser and archived in the audit database.

[0104] Audit Failure Criteria: An audit is considered failed if any of the following occurs: model execution error or timeout leading to forced termination; any CRITICAL (critical defect) rule fails; or the number of MAJOR (major defect) rule failures exceeds a set threshold. The rule engine accurately marks all failed rules as exceptions. The marking information includes: exception rule ID, exception type (parameter out of bounds / incompatible format / running failure, etc.), exception location (specific filename + field name + time), corresponding audit rule description, and system-generated exception reason analysis text. All exception markers are pushed to the browser-side manual review module in a structured list format, triggering the review process.

[0105] S6. After manual review, the review will be revised and resubmitted according to the judgment results, or the model will be modified and resubmitted, or the case will be directly rejected, forming a complete closed loop of "automatic review - anomaly marking - manual review - result feedback".

[0106] Auditors can view the system's precisely marked list of anomalies and the original trial run data in the browser-based manual review module, and perform manual review and judgment, with three processing paths:

[0107] Path A - System Misjudgment: If the reviewer determines that the anomaly is due to a rule setting deviation, the judgment conditions can be modified in the review module for specific rules (such as relaxing the threshold range). After submission, the rule engine will be triggered to partially re-instantiate the modified rules and re-execute the automated review on the model (skipping material parsing and strategy generation, and directly reusing existing running data to re-execute the rule judgment).

[0108] Path B - Actual Problems with the Model: The reviewer confirms the anomaly is due to a problem with the model itself, fills in review comments (explaining the modification requirements for each anomaly), and the system sends the review failure conclusion, a precise list of anomaly markers, and review comments to the model submitter in a notification format. The submitter modifies the anomaly markers and re-uploads the materials, and the system performs a complete automated review of the new batch of materials.

[0109] Path C - Problem Unrectifiable: If the reviewer determines that the model has a fundamental problem that cannot be rectified, the rejection operation will be executed directly. The system will generate a rejection report (including the reason for rejection and all abnormal records), retain the review record, and close the submission task.

[0110] All manual review operations are recorded and stored, including the reviewer's ID, review time, review judgment type, modified content or review comments, and operation records, ensuring that the entire review process is traceable.

[0111] S7. Batch Review and Scheduling: This invention supports automated batch review of hydraulic models. The batch processing mechanism is as follows:

[0112] a. Task queue construction: The system sorts the uploaded models in batches according to model type, priority, and upload time, and constructs an audit task queue;

[0113] b. Concurrent Scheduling: The task scheduling module retrieves tasks from the queue and concurrently schedules multiple models to execute the core processes of 5.1 and 5.2 simultaneously based on the server resource status (the number of concurrent processes is dynamically adjusted according to the server configuration).

[0114] c. Resume execution from the breakpoint: Save the breakpoint state (completed steps, generated rule sets, and completed parsing results) for paused or interrupted audit tasks, and support resuming execution from the breakpoint to avoid duplicate calculations;

[0115] d. Batch Result Summary: After all tasks are completed, the system generates a batch audit summary report, which includes the audit conclusions of each model, pass rate statistics, problem type distribution analysis, etc., and supports export.

[0116] The above is a further description of the present invention in conjunction with the embodiments, and the scope of protection of the present invention is not limited thereto.

Claims

1. An automated review method for water conservancy models based on a rule engine, characterized by: The steps include the following: S1. Construct a basic review rule library for water conservancy models that includes general rules; S2. Rule Instantiation: Receives user-uploaded water conservancy model submission materials, automatically identifies the material type and model application scenario, drives the rule engine to retrieve suitable rules from the rule library, performs fine-grained rule instantiation, and generates a complete set of fine-grained exclusive review rules bound to the model; The fine-grained instantiation of rules includes: a. Condition refinement: Based on the material type and model application scenario, the general condition descriptions in the rule's condition template are refined into specific field-level judgment conditions; b. Parameter assignment: From the system's built-in hydraulic model parameter threshold database, reasonable ranges for each parameter are queried according to the model type and scenario, and parameter placeholders in the basic rules are filled in; c. Logical combination: For cases where the same field involves multiple constraints, multiple sub-rules are combined into a composite rule using AND-OR logic. Perform logical consistency verification on the instantiated exclusive review rule set: check whether there are contradictions between rules, whether there are duplicate rules, and whether the CRITICAL level rules cover all five review dimensions. After the verification is passed, bind and store the exclusive review rule set with the model submission ID. S3. Utilize the large-scale water conservancy model to perform multi-format semantic parsing on the original submitted materials, extract the core parameters, operational boundary conditions, and key performance indicators of the model in a structured manner, and output the parsing results in a standardized format as the driving input for triggering the instantiation of the rule engine; S4. Based on the large model analysis results and the verification indicators in the exclusive audit rule set, automatically generate standardized model operation strategies and drive the automated trial operation of the water conservancy model; Standardized model running strategies include a. Parameter configuration: Map the core parameters and boundary conditions in the analysis results to the parameter format required by the model running environment, and automatically complete the parameter assignment; b. Execution step sequence: In accordance with the hydraulic model operation specifications, a standardized list of model execution steps is generated, with each step including an action description, input dependencies, and timeout limit; c. Verification indicator list: Extract the indicators that need to be monitored in real time from the dedicated audit rule set, generate an indicator list with monitoring thresholds, and connect it with the trial operation module; d. Termination conditions: Define normal termination conditions and forced termination conditions; S5. The rule engine collects real-time data on the trial operation status and verification indicators of the water conservancy model, calls the exclusive audit rule set bound to the model, and executes the rule judgment in parallel according to the audit dimension. If the audit is successful, a standardized audit report is generated; if the audit fails, anomalies are automatically marked and a manual review process is triggered. S6. After manual review, the review will be revised and resubmitted according to the judgment results, or the model will be modified and resubmitted, or the case will be directly rejected, forming a complete closed loop of "automatic review - anomaly marking - manual review - result feedback".

2. The automated review method for water conservancy models based on a rule engine according to claim 1, characterized in that, It also includes batch audit scheduling, and automated batch audit steps for water conservancy models.

3. The automated review method for water conservancy models based on a rule engine according to claim 2, characterized in that, The aforementioned batch automated review specifically includes: task queue construction, concurrent scheduling, breakpoint resume review, and batch result aggregation.

4. The automated review method for water conservancy models based on a rule engine according to claim 1, characterized in that, In step S1, the rules are stored in a JSON structured format and managed according to five dimensions: material completeness, format compliance, parameter rationality, operational effectiveness, and result standardization.

5. The automated review method for water conservancy models based on a rule engine according to claim 4, characterized in that, The rules described in step S1 establish a three-dimensional index based on the audit dimension, material type, and model type.

6. An automated review system for water conservancy models based on a rules engine, comprising a data support layer, a user interaction layer, a core engine layer, and a business execution layer, characterized by: The data support layer stores the basic review rule base for water conservancy models, the water conservancy professional knowledge base, the model material library, and the review database. The user interaction layer includes a material submission and upload module, a review progress and review report viewing module, and a manual review processing module. The core engine layer includes a large-scale water conservancy model engine and a rules engine. The rules engine automatically identifies the material type and model application scenario based on the user-uploaded water conservancy model submission materials. It then drives the rules engine to retrieve suitable rules from the rule library, perform fine-grained rule instantiation, and generate a complete, fine-grained, dedicated review rule set bound to the model. Fine-grained rule instantiation includes: a) Condition refinement: refining the general condition descriptions in the rule's condition template into specific field-level judgment conditions based on the material type and model application scenario; b) Parameter assignment: querying the reasonable range of each parameter from the system's built-in water conservancy model parameter threshold database based on the model type and scenario, and filling the parameter placeholders in the basic rules; c) Logical combination: combining multiple sub-rules into a composite rule using AND / OR logic for cases involving multiple constraints on the same field. Logical consistency verification is performed on the instantiated dedicated review rule set: checking for contradictions between rules, duplicate rules, and whether CRITICAL-level rules cover all five review dimensions. After verification, the dedicated review rule set is bound and stored with the model submission ID. The business execution layer includes a material analysis module, a strategy generation module, a trial operation module, an audit and judgment module, an anomaly marking module, and a review and management module; The material analysis module uses the large hydraulic model to perform multi-format semantic analysis on the original submitted materials, extracts the core parameters of the model, the boundary conditions of operation, and the key indicators of the results in a structured manner, and outputs the analysis results in a standardized format as the driving input to trigger the instantiation of the rule engine. The strategy generation module automatically generates standardized model operation strategies based on the large model analysis results and verification indicators in a dedicated set of audit rules; the trial operation module is responsible for the automated trial operation of the hydraulic model; the standardized model operation strategies include... a. Parameter configuration: Map the core parameters and boundary conditions in the analysis results to the parameter format required by the model running environment, and automatically complete the parameter assignment; b. Execution step sequence: In accordance with the hydraulic model operation specifications, a standardized list of model execution steps is generated, with each step including an action description, input dependencies, and timeout limit; c. Verification indicator list: Extract the indicators that need to be monitored in real time from the dedicated audit rule set, generate an indicator list with monitoring thresholds, and connect it with the trial operation module; d. Termination conditions: Define normal termination conditions and forced termination conditions; The review and judgment module collects real-time data on the trial operation status and verification indicators of the water conservancy model from the rule engine, calls the exclusive review rule set bound to the model, and performs rule judgments in parallel according to the review dimensions. If the review is successful, a standardized review report is generated. The anomaly marking module automatically marks anomalies for failed reviews and triggers a manual review process. The review management module is responsible for manual review. After manual review, it will correct and re-examine the results according to the rules, modify the model and re-examine, or directly reject the review, forming a complete review closed loop of "automatic review - anomaly marking - manual review - result feedback".

7. The automated review system for water conservancy models based on a rule engine according to claim 6, characterized in that, It also includes a batch review and scheduling module, which performs batch automated review of water conservancy models, completes task queue construction, concurrent scheduling, breakpoint resume review, and batch result summary.

Images