Engineering regulation generation and verification method and system based on interactive logic atlas

Abstract

The invention discloses an engineering regulation generation and verification method and system based on an interactive logic map, and the method achieves the complete transparency, traceability and verifiability of an engineering regulation generation and review process, and effectively solves the problems that in a traditional method, calculation is black, errors are difficult to trace, and review is low in efficiency. Accuracy and safety of regulation compiling are improved, and a trust mechanism of an engineer to an AI generation result is established.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of human-computer interaction and artificial intelligence application, in particular to visual modeling, knowledge graph, data traceability technology, and especially to an engineering regulation generation and verification method and system based on an interactive logic graph. BACKGROUND

[0002] In the field of safety-related industries such as mining, the compilation and review of engineering regulations is a very rigorous task. Its definition is: according to a series of industry standards, physical formulas, field survey data and expert decisions, a legal document is generated which is a guide for on-site operations, has a rigorous logic and accurate data. There are many pain points in the generation and review of traditional regulations.

[0003] The existing technology mainly solves such problems in the following ways: pure manual compilation and review: engineers manually review standards, apply formulas, use calculators or spreadsheets for calculations, and finally fill the results into document templates. Although this method gives engineers complete control, its shortcomings are obvious: low efficiency, prone to errors (such as data transcription errors, misuse of formulas), and the calculation process is not transparent, making it difficult to trace the complete calculation chain of a certain data during review, resulting in a huge workload and low reliability. Calculation software based on fixed rules: these software solidify standards and formulas in program code, and users can get results by inputting parameters. Its main disadvantage is the "black box" feature. Users cannot directly see the intermediate process and logical dependency of the calculation. If the input data is incorrect, or a certain standard built into the software is not updated in time, the error will silently propagate to the final result, and users will have difficulty finding and locating the problem source. Emerging general artificial intelligence generation (such as LLM): in recent years, large language models have shown strong text generation capabilities, and in theory, they can generate draft regulations. However, its core defect lies in its "black box" and "unreliable" nature. The model may produce "illusion" data that appears reasonable but is actually incorrect, and its internal logic cannot be verified and audited. In the high-risk engineering field, it is unacceptable to completely entrust decision-making to an AI that cannot explain its reasoning process.

[0004] In summary, the existing technology has not effectively solved the core contradiction in the engineering regulation generation process: how to improve efficiency while ensuring complete transparency, traceability, and reviewability of the process, thereby establishing trust in machines by humans. There is an urgent need for a method that "whitewashes" the AI calculation process and allows experts to intervene, verify, and deduce efficiently in a loop. SUMMARY

[0005] The main purpose of the present application is to provide an engineering regulation generation and verification method based on an interactive logic graph.

[0006] Another object of the present application is to provide an interactive logic graph-based engineering regulation generation and verification system.

[0007] To achieve the above object, the first aspect of the present application provides an interactive logic graph-based engineering regulation generation and verification method, comprising: S1, parsing the to-be-inspected engineering regulation document and generating a status graph A reflecting the calculation logic thereof, and simultaneously constructing a standard graph B based on the initial conditions and the latest industry standards extracted from the document; S2, performing structural, rule and numerical difference analysis on the status graph A and the standard graph B through a multi-dimensional comparison algorithm, identifying calculation path loss, logic rule error or parameter calculation deviation; S3, generating an interactive review report containing a list of difference points, wherein each difference is associated with the corresponding node of the status graph A and the standard graph B, and the instant positioning of problem tracing is realized through hyperlinks; S4, allowing the user to modify the parameters or rules of the status graph A in the review report, triggering instant recalculation of the dependent relationship chain and updating the graph state, so as to verify whether the corrected calculation logic meets the constraint conditions of the standard graph B.

[0008] In an embodiment of the present application, the S1 of parsing the to-be-inspected engineering regulation document and generating a status graph A reflecting the calculation logic thereof further comprises: S11, using optical character recognition (OCR) technology to process scanned PDF or image format documents, converting them into editable text streams, and extracting engineering parameter nodes including roadway cross section, gas emission quantity and number of workers through named entity recognition (NER) model in natural language processing technology; S12, based on the context information in the document, using relation extraction (RE) model to identify key phrases, constructing directed edges between parameter nodes, and forming a directed acyclic graph (DAG) structure reflecting the calculation logic of the original document.

[0009] In an embodiment of the present application, the S2 of performing structural, rule and numerical difference analysis on the status graph A and the standard graph B through a multi-dimensional comparison algorithm further comprises: S21, performing topological alignment on structural differences, detecting missing nodes by traversing the node sets of the status graph A and the standard graph B, and generating corresponding problem descriptions.

[0010] In an embodiment of the present application, the S3 of generating an interactive review report containing a list of difference points further comprises: S31, each difference point is bound to the corresponding node of the status atlas A and the standard atlas B, and instant positioning of problem tracing is realized through a hyperlink in HTML format, which displays the complete calculation path and parameter source of the difference node in the two atlases after being clicked.

[0011] In an embodiment of the present application, the S4 allows the user to modify the parameters or rules of the status atlas A in the review report, and further comprises: S41, when the user modifies the parameters, the system marks the node as manually covered by the user, and triggers instant recalculation of the downstream nodes based on the dependency relationship chain, and the recalculation process comprises:

[0012]

[0013]

[0014]

[0015] wherein, is the gas emission amount, is the gas unevenness coefficient, is the allowable gas concentration, is the number of workers, is the air volume per person, is the minimum air speed, is the roadway cross section.

[0016] To achieve the above purpose, the second aspect embodiment of the present application proposes an engineering regulation generation and verification system based on an interactive logical atlas, comprising: a document analysis and atlas generation module, which is used for analyzing the to-be-reviewed engineering regulation document and generating a status atlas A reflecting the calculation logic thereof, and simultaneously independently constructing a standard atlas B based on the initial conditions and the latest industry standards extracted from the document; A multi-dimensional comparative analysis module is used for analyzing the structural, rule and numerical differences between the status atlas A and the standard atlas B through a multi-dimensional comparative algorithm, identifying calculation path loss, logical rule error or parameter calculation deviation; An interactive review report generation module is used for generating an interactive review report containing a difference point list, wherein each difference is associated with the corresponding node of the status atlas A and the standard atlas B, and instant positioning of problem tracing is realized through a hyperlink; A parameter modification and dependency recalculation module is used for allowing the user to modify the parameters or rules of the status atlas A in the review report, triggering instant recalculation of the dependency relationship chain and updating the atlas state, so as to verify whether the corrected calculation logic meets the constraint conditions of the standard atlas B.

[0017] The method and system for generating and verifying engineering regulations based on an interactive logic graph according to the embodiments of the present application aim to overcome the fundamental defects of "black box" calculation, process opacity, and difficulty in tracing and reviewing in the prior art. The present application aims to provide a brand-new method and system for generating and verifying engineering regulations based on an interactive logic graph. By visualizing the complex calculation and decision-making process in real time and dynamically as a logic graph, the present application makes the "thinking process" of AI explicit, achieving complete transparency and auditability of the process. More importantly, it creates a new paradigm of human-machine collaboration of "AI paving the way and experts checking", allowing engineers to review, modify parameters, and recalculate instantly at any node on the graph, thereby performing rapid risk deduction and scheme optimization, and fundamentally solving the AI trust problem in high-risk fields. BRIEF DESCRIPTION OF DRAWINGS

[0018] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which: Figure 1 A flowchart of a method for generating and verifying engineering regulations based on an interactive logic graph according to an embodiment of the present application is provided. Figure 2 A flowchart of a review mode and a generation mode according to an embodiment of the present application is provided. Figure 3 A structural schematic diagram of a system for generating and verifying engineering regulations based on an interactive logic graph according to an embodiment of the present application is provided. DETAILED DESCRIPTION

[0019] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0020] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings and in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0021] A method and system for generating and verifying engineering regulations based on an interactive logic graph according to an embodiment of the present application will be described below with reference to the accompanying drawings.

[0022] Embodiment 1 The present embodiment provides a method for generating and verifying engineering regulations based on an interactive logic graph. As shown in FIG. 1, the method comprises the following steps. Figure 1As shown, the method includes the following steps: S1 parses the engineering specification document to be reviewed and generates a current status map A that reflects its calculation logic. At the same time, it independently constructs a standard map B based on the initial conditions extracted from the document and the latest industry standards.

[0023] Specifically, the system first performs structured processing on the input documents to be reviewed using multimodal document parsing technology. For scanned PDFs or image-formatted documents, the system employs a high-precision OCR engine (such as Tesseract OCR or Baidu OCR) to recognize text, converting unstructured text into a processable character stream. For electronic documents (such as .docx, .pdf, etc.), a document parser extracts the text content, preserving structural information such as chapters, tables, and formulas. Subsequently, the system utilizes Natural Language Processing (NLP) technology, particularly Named Entity Recognition (NER) and Relation Extraction (RE) models, to perform semantic parsing on the text, identifying key information such as engineering parameters, computational nodes, and standard references.

[0024] During parameter extraction, the system binds numerical information (such as tunnel cross-section, gas emission rate, wind speed, etc.) in the document with their corresponding units, forming standardized input nodes. For example, from "Tunnel Cross-section: The node `[tunnel cross-section: 10 m²]` is extracted. At the same time, the system infers the logical dependencies between parameters through context analysis and keyword matching (such as "calculated from...", "take the maximum value", etc.), constructs a directed acyclic graph (DAG) structure, and forms "current status map A".

[0025] When constructing the "Standard Map B", the system uses initial conditions extracted from documents and calls upon a built-in engineering knowledge base. This knowledge base contains the latest national / industry standards (such as the "Coal Mine Safety Regulations"), calculation formulas (such as $Q_{\mathrm{gas}} = (100 \times q_{\mathrm{gas}} \times K) / C_{\mathrm{permitted}}$), and parameter lookup tables (such as the K coefficient table, per capita air demand table, etc.). Following standard rules, the system independently executes the calculation process, generating calculation nodes and decision nodes that conform to current specifications, thus constructing the "Standard Map B".

[0026] The technical indicators of this step include OCR recognition accuracy (≥95%), NER model recall rate (≥90%), relationship extraction F1 value (≥85%), and the integrity and consistency of the graph construction. In application scenarios, this step is widely used in high-reliability engineering fields such as mine ventilation, building safety, and power systems, supporting automatic review and compliance verification of existing regulations. The technical value lies in realizing the logical explicitness of document content, laying a structured foundation for subsequent graph comparison and review report generation, thereby improving review efficiency and accuracy and ensuring the compliance and safety of engineering regulations.

[0027] Further, S1 includes: S11, using optical character recognition (OCR) technology to process scanned PDF or image format documents, converting them into editable text streams, and extracting engineering parameter nodes through named entity recognition (NER) models in natural language processing (NLP) technology, including roadway cross-section, gas emission, and number of workers.

[0028] Specifically, in some implementations, this step uses optical character recognition (OCR) technology to extract text from scanned PDF or image format engineering regulation documents, and extracts key engineering parameter nodes through named entity recognition (NER) models in natural language processing (NLP). This process is the basis for constructing the "current graph A" in the entire review mode, and the technical implementation needs to consider the accuracy of text recognition and the standardization of parameter extraction.

[0029] OCR processing usually uses deep learning-based text detection and recognition models such as EAST, CRNN, or PaddleOCR to achieve high-precision extraction of text in unstructured images or scanned documents. The OCR engine first preprocesses the document image, including grayscale, binarization, denoising, and tilt correction, to improve recognition quality. The recognized text is output in a structured format (such as JSON or XML), including character content, position coordinates, font size, and other meta-information, for subsequent processing.

[0030] Further, the system performs entity recognition on the OCR-extracted text stream using a NER model. This model can be fine-tuned based on pre-trained language models like BERT, RoBERTa, etc., to adapt to the specific terminology and expression patterns in engineering texts. The engineering parameter nodes identified by the NER model include but are not limited to roadway cross-section, gas emission, number of workers, etc., and their numerical formats must conform to the regular expression `[0-9A-Z]{18}` to ensure data standardization and parsability. For example, the system can identify "roadway cross-section: 10 m²" as an entity type with the value "10" and verify whether it meets the pre-set format requirements.

[0031] The accuracy of OCR recognition should generally reach more than 95% to ensure the integrity of key data. The F1 value of the NER model should not be less than 0.92 to ensure the accuracy and recall rate of parameter extraction. In addition, the system supports multi-language recognition and entity tagging, suitable for mixed scenarios of Chinese, English, and other engineering documents.

[0032] This step is widely used in fields with high reliability requirements such as mine ventilation, building safety, and power engineering. For example, in the review of mine ventilation regulations, the system can automatically identify key parameters such as roadway cross-section and gas emission and map them as nodes in the graph, providing a data foundation for subsequent logical deduction and rule verification.

[0033] The technical effect of this step is to achieve efficient conversion from unstructured documents to structured parameter nodes, providing data support for building interactive logical graphs. Through the joint application of OCR and NER, the system can automatically extract and verify engineering parameters, significantly improving the automation level and data reliability of regulation review.

[0034] S12, based on the context information in the document, a relationship extraction (RE) model is used to identify key phrases and construct directed edges between parameter nodes, forming a directed acyclic graph (DAG) structure reflecting the original document's calculation logic.

[0035] Specifically, in some implementations, the system identifies key phrases (such as "because... therefore..." and "according to... calculated") in the document through a relationship extraction (RE) model and constructs directed edges between parameter nodes based on this, forming a directed acyclic graph (DAG) structure reflecting the original document's calculation logic. This step is the core link of the entire "current graph A" construction process, and its technical implementation is based on the deep integration of natural language processing (NLP) and graph structure modeling.

[0036] The system first uses a deep learning-based relation extraction model, such as BERT-BiLSTM-CRF or a multi-head attention mechanism model based on Transformer, to perform semantic analysis on the text in the document. This model, after pre-training and fine-tuning on an engineering regulation corpus, can identify the logical dependency relationships between parameter nodes. For example, when the system identifies the sentence "According to the gas emission amount, the required air volume is calculated as ", the RE model will extract the "according to... calculated" relationship between "gas emission amount" and "required air volume", and establish a directed edge from the "gas emission amount" node to the "required air volume" node in the graph. This edge not only represents data dependency, but also implies the source and direction of the calculation logic.

[0037] The input of the relation extraction model is the text passage extracted from the document, and the output is the node pair and its relationship type. The accuracy and F1 score of the model usually need to reach more than 90% to ensure the reliability of the graph construction. In addition, the system supports multiple relationship types, such as "according to", "deduce", "take the maximum value", "take the minimum value", etc., each corresponding to different graph edge attributes, facilitating subsequent rule-based comparative analysis.

[0038] This step is widely used in high-reliability engineering fields such as mine ventilation, building safety, and power systems. For example, in the review of mine ventilation regulations, the system can automatically identify the logical relationships between parameters such as "roadway cross section", "gas emission amount", and "number of people", and construct a complete calculation chain. Engineers can visually view the source of each parameter and its role in the overall logic through the graph, thereby achieving a comprehensive understanding and verification of the document calculation process.

[0039] Further, the technical effect of this step is that it explicitly displays the calculation logic that was originally implicit in the text as a graph structure, achieving visualization and traceability of data flow. By constructing a DAG, the system not only identifies the dependency relationships between parameters, but also detects circular dependencies or logical breaks, thereby improving the structural integrity and logical consistency of the regulation document. This technology provides a solid data foundation for subsequent graph comparison, review report generation, and risk deduction, and is a key link in realizing "white-box" engineering regulation generation and verification.

[0040] S2, through a multi-dimensional comparison algorithm, analyzes the structural, rule, and numerical differences between the current graph A and the standard graph B, identifying missing calculation paths, logical rule errors, or parameter calculation deviations.

[0041] The system first aligns the nodes and edges of the two graphs using a graph matching algorithm (such as maximum common subgraph matching, node attribute similarity matching, etc.). Node alignment is based on semantic labels (such as "calculate air volume according to gas" and "calculate air volume according to number of people") and parameter attributes (such as units and numerical ranges). Edge alignment is based on semantic consistency of dependencies (such as "calculate according to..." and "take the maximum value"). After alignment, the system performs structural, rule, and numerical difference analysis in turn.

[0042] In structural difference analysis, the system detects whether a calculation node or logical branch that should exist in graph B exists in graph A. For example, if the "calculate air volume according to minimum wind speed" node exists in graph B but is missing in graph A, it is determined to be a structural defect. In rule difference analysis, the system compares the calculation logic within the nodes, such as formula expressions, function call methods, and parameter sources. For example, if an incorrect decision function is used in graph A (such as Average() instead of Max()), it is determined to be a rule error. In numerical difference analysis, the system compares the output results of the nodes under the premise of structural and rule consistency. If the numerical deviation exceeds a pre-set threshold (such as 5%), it is determined to be a numerical difference.

[0043] The system supports multiple comparison modes, including exact matching, fuzzy matching, and version control matching. For example, version control of parameter K can be based on standard update timestamps. If graph A references K=1.2 in the 2015 version of the standard, while graph B uses K=1.5 in the 2020 version of the standard, the system can identify it as a rule difference. In addition, the system supports quantitative evaluation of calculation errors, such as using the relative error formula to determine whether the numerical deviation is within the allowed range.

[0044] This step is widely used in engineering fields with high reliability requirements such as mine ventilation, building safety, and power systems. For example, in the review of mine ventilation regulations, the system can automatically identify missing wind speed calculation items or incorrect gas coefficient references in old regulations, thereby assisting engineers in quickly locating the root cause of the problem and improving review efficiency and accuracy.

[0045] The technical effect of this step is that through multi-dimensional comparison of structure, rules, and values, the system can achieve automatic and traceable review of engineering regulation documents, significantly improving the comprehensiveness and credibility of the review, and providing key support for the human-machine collaboration mode of "AI paving the road and experts checking the road".

[0046] Further, S2 includes: S21, topological alignment of structural differences, by traversing the node sets of the existing graph A and the standard graph B, detecting missing nodes and generating corresponding problem descriptions.

[0047] Specifically, in some implementations, topological alignment of structural differences is a key step in the atlas comparison engine of the present invention, and the core goal is to identify the deviation of the current atlas A and the standard atlas B in terms of logical integrity through graph structure matching and difference detection. The technical implementation of this step is based on the topological alignment algorithm in graph theory, combined with node matching and path tracking mechanism, to ensure the comparability of the two atlases in the structure dimension.

[0048] In terms of specific operation, the system first traverses the node sets of the current atlas A and the standard atlas B, and adopts a bidirectional matching strategy based on node labels (Node Label) and dependency relationships (Edge Relation). Node labels usually include parameter names (such as "tunnel cross section", "gas emission") and node types (input nodes, calculation nodes, decision nodes, etc.), while dependency relationships are composed of edge directionality and semantic relationships (such as "calculate according to", "take maximum value"). In the matching process, the system uses graph isomorphism detection algorithms (such as Weisfeiler-Lehman algorithm) to align the structures of the two atlases, and identifies the missing nodes in A or the necessary nodes in B that are not referenced by A.

[0049] For example, when the system detects that the "calculate air volume according to minimum wind speed" node is missing in the current atlas A, it will output a prompt message "

structural defect

[0050] Further, the implementation of this step relies on the standardized processing of node identifiers (Node ID) and semantic labels in the atlas. In the standard atlas B, each node is semantically annotated according to the ISO / IEC 24615 standard, ensuring that it can be uniquely identified and matched. The system constructs a node mapping table (Node Mapping Table) to one-to-one correspond the nodes in A and B, and if no match is found, it is determined as a structural difference.

[0051] In practical applications, this step is widely used in high-reliability engineering fields such as mine ventilation, building safety, and power systems, especially when reviewing old regulations or generating new regulations, to ensure that all necessary calculation links are included, thereby improving the compliance and safety of the regulations. Its technical value lies in the comprehensive coverage of engineering logic chains through structural integrity detection, avoiding systemic risks caused by missing key nodes.

[0052] S3 generates an interactive review report containing a list of discrepancies, where each discrepancy is associated with a corresponding node in the current status map A and the standard map B, and the source of the problem is located instantly through hyperlinks.

[0053] Specifically, the system first structures the list of differences output by the graph comparison engine. Each difference is treated as an independent review item, containing the issue type (e.g., structural difference, regularity difference, numerical difference), issue description, involved nodes, original document snapshot, and standard basis. Each item is linked via an HTML anchor (` The label is bidirectionally linked to the specific node in the graph. For example, when a user clicks on a certain difference description in the report, the front-end page will dynamically load and display a side-by-side view of the two graphs, and highlight the difference node and its dependency path using CSS animation (such as flashing highlight), facilitating the user to quickly identify the root cause of the problem.

[0054] The system supports multiple difference type identifications, such as "structural defect", "rule error", "numerical deviation", etc., and defines a standardized evaluation template for each type of difference. For example, for numerical differences, the system will calculate the relative error and automatically mark it as a high-risk item according to the preset threshold (e.g. error exceeding 5%). In addition, the system also supports labeling the version information of the node, such as whether the K coefficient is the old version standard (1.2) or the new version standard (1.5), to assist the reviewer in judging the impact of standard updates.

[0055] This interactive review report is widely used in high-reliability engineering fields such as mine ventilation, building safety, and power systems. For example, in the review of mine ventilation systems, engineers can quickly locate the numerical deviation of the "gas-calculated air volume" node and trace whether it is caused by K coefficient version error or input data transcription error. Through hyperlink jumping, engineers can directly view the comparison of the original calculation path and the standard calculation path in the graph, and thus make accurate corrections.

[0056] This step realizes the "visual closed loop" of the review process, converting the logical errors that are difficult to trace in traditional document review into interactive and locatable graph differences. Through immediate positioning and tracing, the system significantly reduces the workload of manual review and improves the accuracy and auditability of the review, providing an efficient, transparent, and reliable intelligent review mechanism for high-risk engineering fields.

[0057] Further, S3 comprises: S31, binding each difference point to the corresponding node of the existing graph A and the standard graph B, realizing immediate positioning of problem tracing through HTML format hyperlinks, which, when clicked, display the complete calculation path and parameter source of the difference node in the two graphs side by side.

[0058] Specifically, in some implementations, the step "binding each difference point to the corresponding node of the existing graph A and the standard graph B, realizing immediate positioning of problem tracing through HTML format hyperlinks" is the key link in the review mode of the present application to realize "white-boxing" and "traceability". This step builds semantic mapping relationships between nodes and embeds hyperlinks in the review report, enabling users to quickly locate and compare the complete calculation path and parameter source of the difference node in the two graphs, thereby realizing accurate review and risk identification of engineering regulations.

[0059] The system first performs semantic alignment of the nodes in the current graph A and the standard graph B based on the list of difference points output by the graph comparison engine. The alignment process relies on the naming rules, parameter types, calculation formulas, and contextual logical relationships of the nodes. For example, if there is a node named "Calculate air volume by gas" in the "current graph A", and there is a node with the same semantics in the "standard graph B", the system will bind them as a pair of difference nodes. After binding, the system generates a hyperlink element in the HTML review report for each difference point, with the anchor pointing to the corresponding node position in the two graphs.

[0060] The generation of hyperlinks must meet the following standards: the difference point must contain the node ID, node type (input node, calculation node, or decision node), calculation formula, input parameters and their sources (such as original documents, knowledge base entries, or user input), output results and their units. The system encapsulates parameters and formulas as ` Tags, through JavaScript event binding, realize the parallel display of the complete calculation path of the node in the two graphs after clicking.

[0061] This step is widely used in mine ventilation, power system, building safety and other high reliability engineering fields. For example, when reviewing a mine ventilation regulation, if it is found that the "air volume calculated by the number of people" node is missing in the current graph, the system will generate a hyperlink that, when clicked, will display the calculation path of the node in the standard graph and the basis for the "XXth article of the Coal Mine Safety Regulations", helping engineers quickly identify structural defects and make corrections.

[0062] This step realizes "instant traceability" and "visual comparison" of the review process, significantly improving review efficiency and accuracy. Through the hyperlink mechanism, users do not need to manually search for documents or graphs to directly locate the problem node and view its complete calculation chain, effectively avoiding misjudgment and omissions caused by scattered information and opaque logic in traditional reviews. Further, this step provides data consistency protection for subsequent "one-click generation", ensuring that the final output regulation document is fully aligned with the standard logic.

[0063] S4, allows the user to modify the parameters or rules of the current graph A in the review report, triggers the instant recalculation of the dependency relationship chain and updates the graph state, to verify whether the modified calculation logic meets the constraints of the standard graph B.

[0064] Specifically, in some implementations, the step "allowing the user to modify the parameters or rules of the current graph A in the review report, triggering the instant recalculation of the dependency relationship chain and updating the graph state, to verify whether the modified calculation logic meets the constraints of the standard graph B" is the core mechanism of the "white-box" review and "expert-in-the-loop" human-machine collaboration in this invention. This step realizes real-time verification and risk deduction of the calculation logic in the engineering regulation by building a dynamic response mechanism for the graph nodes.

[0065] The system provides an interactive entry for each difference item in the review report, and the user can click to enter the corresponding node in "current graph A" to view its current calculation formula, input parameters and their sources. After the user modifies the node parameters or rules, the system identifies the downstream dependency path of the node in the graph through the dependency relationship tracking algorithm in the graph engine, and triggers chain recalculation. The recalculation process uses a topological sorting algorithm for directed acyclic graphs (DAG) to ensure that the calculation order conforms to the logical dependency relationship, avoiding system instability caused by circular calculation. After the recalculation is completed, the system synchronizes the updated node state to graph A and compares it with the corresponding node in graph B in real time to determine whether it meets the standard constraints.

[0066] The system supports marking and checking the numerical accuracy, unit system, and source type (such as user input, knowledge base reference, formula derivation) of node parameters. For example, the unit of the gas emission amount node is , and the input value must satisfy , and must be consistent with the parameters of the roadway cross section , wind speed , etc. The built-in verification module in the system will perform compliance checking on the modified parameters according to the rules in standard atlas B to ensure that they meet the current industry standards such as the "Coal Mine Safety Regulations".

[0067] This step is widely used in high-reliability engineering fields such as mine ventilation, shaft and tunnel design, and safety assessment. For example, when engineers review old regulations, they find that the K coefficient of the "gas-calculated air volume" node references the old standard (1.2), while the new standard requires 1.5. By modifying the parameter, the system immediately triggers a recalculation, updates the air volume result, and compares it with the data in standard atlas B to determine whether it meets the safety threshold requirement of .

[0068] The technical effect of this step is to achieve dynamic verification and closed-loop feedback in the engineering regulation review process. Through immediate recalculation and atlas comparison, users can quickly assess the impact of parameter modification on the overall logic without regenerating the entire document, thereby improving review efficiency, reducing the risk of misjudgment, and enhancing trust in AI-generated results.

[0069] Further, S4 includes: S41, when the user modifies the parameter, the system marks the node as user-manual override, and triggers immediate recalculation of downstream nodes based on the dependency relationship chain, the recalculation process includes:

[0070]

[0071]

[0072]

[0073] wherein, is the gas emission amount, is the gas unevenness coefficient, is the allowable gas concentration, is the number of workers, is the per capita air volume requirement, is the minimum air speed, is the roadway cross section.

[0074] Specifically, when a user modifies the parameters of a certain node in the interactive logical graph, the system marks the node as "user manual override" and triggers the instant recalculation of downstream nodes based on the predefined dependency chain in the graph. This step is the core of the "AI paving, expert checking" human-machine collaboration mechanism, and its technical implementation relies on the topology analysis of the graph, the parameter change tracking mechanism, and the efficient computing engine.

[0075] The system first identifies all downstream dependent nodes of the node through the directed acyclic graph (DAG) of the graph. Each node carries metadata of its dependency when it is constructed, and the system traverses all nodes affected by the modified node through depth-first search (DFS) or breadth-first search (BFS) algorithm. Once the nodes to be recalculated are identified, the system will call the built-in computing engine to perform the calculation again according to the formula or decision rule bound to the node. For example, if the user modifies the value of the "gas emission" node, the system will recalculate the output of the "calculate air volume by gas" node according to the formula and update the result to the graph.

[0076] The system supports real-time modification of multiple data types, including but not limited to floating-point numbers, integers, percentages, and unit dimensions. The modification operation must meet certain input validation mechanisms, such as unit consistency check, numerical range limitation (such as wind speed cannot be lower than ), standard version compatibility verification, etc. In addition, the system records the timestamp, operator identity, and parameter values before and after each modification to support subsequent version backtracking and responsibility tracking.

[0077] This step is widely used in high-reliability engineering fields such as mine ventilation, safety assessment, and equipment selection. For example, in mine ventilation design, engineers can temporarily modify the roadway section or gas emission, and the system will immediately feedback the impact on key nodes such as total air volume and fan selection, thereby supporting "What-If" analysis and risk deduction.

[0078] The technical effect of this step is to realize the dynamic response and visual feedback of the engineering calculation process, enhancing the interactivity and real-time performance of the system. Through the "user manual override" mechanism, the system can clearly distinguish the data sources of AI automatic generation and manual intervention, ensuring that the final generated procedure document is logically interpretable and data verifiable, thereby improving the transparency and reliability of engineering decisions.

[0079] The engineering procedure verification method based on the interactive logical graph of the embodiment of the application realizes the complete visualization and traceability of the engineering procedure generation and review process, improves the calculation transparency and human-machine collaboration efficiency, and ensures the accuracy and consistency of the procedure logic and data.

[0080] Further, it also includes: S5, version check the difference points in the review report with the latest industry standards in the knowledge base, if the reference standard version is found to be outdated, generate prompt information, and provide version comparison view of standard clauses.

[0081] Specifically, in the review mode, the graph comparison engine is responsible for multi-dimensional difference analysis of "current graph A" and "standard graph B". The core goal is to identify possible structural deficiencies, rule errors and numerical deviations in the document, so as to provide accurate difference point data for subsequent generation of review report. The technical implementation of this step is based on graph structure alignment and difference detection algorithm, combined with industry standard version management mechanism, to ensure the authority and traceability of the review results.

[0082] In some implementations, the system first performs semantic matching and topological alignment of the nodes of the two graphs. Node matching uses a method based on named entity recognition (NER) and semantic similarity calculation to one-to-one map the parameter nodes in "current graph A" to the corresponding calculation nodes in "standard graph B". For example, if there is a "calculate air volume by gas" node in "current graph A", the system will automatically find the node with the same semantics in "standard graph B" and establish a corresponding relationship. If it cannot be matched, it is determined as a structural difference, such as missing necessary calculation nodes or introducing redundant nodes.

[0083] Further, the system performs version check on the calculation rules inside the nodes. In rule difference detection, the system compares the industry standard version referenced by the node with the latest version stored in the knowledge base. For example, if a node in "current graph A" references K coefficient as 1.2 in "Coal Mine Safety Regulations" 2015 version, while the latest version in the knowledge base is 2023 version, in which K coefficient has been updated to 1.5, the system will trigger the version check mechanism and generate a prompt information: "

standard version risk

[0084] Optionally, the system also provides a version comparison view of standard clauses in the review report, showing the definition, value range and application conditions of related parameters in 2015 version and 2023 version in the form of table or parallel text, so as to facilitate engineers to quickly identify changes and evaluate their impact on calculation results. The view supports interactive expansion and folding, improving the review efficiency.

[0085] This step plays a crucial role in the overall technical solution. It not only automates the review of engineering specification documents but also ensures, through a version verification mechanism, that all calculations are based on the latest and compliant industry standards, thereby effectively mitigating security risks caused by outdated standards. Its technical value lies in improving the accuracy of the review, enhancing standard compliance, reducing the cost of manual verification, and providing auditable and traceable intelligent review capabilities for high-reliability engineering fields.

[0086] The engineering procedure generation and verification method based on interactive logic graphs in this invention verifies the differences in the review report against the latest industry standards in the knowledge base, and automatically generates version risk warnings and clause comparison views when outdated standard references are detected. This further improves the standardization and timeliness of procedure review and effectively avoids technical compliance risks caused by outdated standard versions.

[0087] Example 2 This invention proposes a method for generating and verifying engineering specifications based on interactive logic graphs. The overall process is as follows: Figure 2 As shown, this method transforms the traditional linear, static document generation process into a dynamic, visual, and interactive graph construction and verification process. It mainly includes four core parts: dynamic construction of the logical graph, transparent interaction at the node level, real-time recalculation of the sandbox simulation, and one-click generation and intelligent review.

[0088] In one embodiment of the present invention, such as Figure 2 The review mode shown here: This mode performs automated and traceable intelligent review of existing engineering specification documents. Its detailed technical process is as follows: Step 1: Input the document to be reviewed and perform format adaptation and content extraction: The system receives the engineering specification document uploaded by the user, supporting file formats including .doc, .docx, .pdf, .wps, etc. For scanned PDFs or documents existing in image form, the system first uses Optical Character Recognition (OCR) technology to convert them into an editable and analyzable text stream; for native electronic documents (such as Word or WPS files), it directly parses and extracts the text content. During this process, the system simultaneously recognizes the document's chapter structure, paragraphs, tables, formulas, and key engineering statements. For example, given a document titled "Design Code for Mine Ventilation Systems," the system can accurately extract complete information including engineering parameters, references, and calculation conclusions, such as "Roadway cross-section: 10m²," "Absolute gas emission rate at the working face is 5m³ / min," "According to Article XX of the 'Coal Mine Safety Regulations,' the required air volume based on the gas emission rate is 500m³ / min," and "The final determined total air volume is 600m³ / min." This information is then output as structured raw text data, providing a foundation for subsequent analysis. Step two: Analyzing document logic, automatically generating "Current Atlas A": Based on the structured text output in the first step, the system calls natural language processing models, especially named entity recognition (NER) and relation extraction (RE) techniques, to automatically identify and extract key engineering parameters and their values in the document, forming the "nodes" of the atlas. For example, extract atomic information nodes: [Tunnel cross-section: 10m²], [Gas emission: 5m³ / min], [Gas-calculated air volume: 500m³ / min], [Total air volume: 600m³ / min], etc. At the same time, the system reversely infers the dependency relationships and calculation paths between these parameters according to the logical conjunctions in the context (such as "according to… calculated", "because… so…" "take the maximum value in…" etc.), constructing directed "edges". For example, from "the required air volume calculated according to gas emission is 500m³ / min", it is inferred that [Gas emission] → [Gas-calculated air volume] calculation dependency relationship. Finally, the system integrates all nodes and edges to generate a directed acyclic graph (DAG) form "Current Atlas A", which completely and structurally restores the data flow, calculation logic and decision basis implied in the original document.

[0089] Step three: Based on initial conditions, independent calculation, generate "Standard Atlas B": The system automatically identifies and extracts initial conditions for engineering calculation from "Current Atlas A" as input, such as [Tunnel cross-section: 10m²], [Tunnel length: 800m], [Absolute gas emission: 5m³ / min], [Number of workers: 50 people], etc. These initial parameters are sent to a separate calculation engine embedded with industry standards (such as "Coal Mine Safety Regulations") and standard algorithms. This engine does not rely on the original document's calculation process, but strictly follows the current technical standards to re-execute the complete process from air volume calculation, wind resistance checking to final air volume determination. The calculation results are organized into a directed acyclic graph aligned with the structure of "Current Atlas A", i.e. "Standard Atlas B". This atlas represents the correct calculation path and result under the same initial conditions according to the standard, which is used for subsequent automatic comparison with "Current Atlas A" node by node and logic by logic, thereby realizing automatic, traceable and high-precision compliance review of the regulation document.

[0090] In an embodiment of the present application, a knowledge base is called: the system accesses the built-in engineering knowledge base, which stores the latest national / industry standards, safety regulations, calculation formulas, query tables, etc. Standardized calculation: According to the standard rules in the knowledge base, a complete calculation and decision is independently performed.

[0091] Example 1: Calculate the air volume according to the gas emission (calculate node): Formula: Q_gas = (100 * q_gas * K) / C_allow; Parameters: q_gas (gas emission) = 5 m³ / min; K (gas emission unevenness air volume coefficient) = 1.5 (obtained by checking the standard table); C_allow (allowed gas concentration) = 1.0% (obtained by checking the standard table). Calculation: Q_gas = (100 * 5 * 1.5) / 1.0 = 750 m³ / min.

[0092] Example 2: Calculate the air volume according to the number of people (calculate node): Formula: Q_people = N * q_people; Parameters: N (number of simultaneous workers) = 50 people; q_people (air volume per person) = 4 m³ / min (obtained by checking the standard table). Calculation: Q_people = 50 * 4 = 200 m³ / min.

[0093] Example 3: Calculate the air volume according to the minimum air speed (calculate node): Formula: Q_speed = v_min * S * 60; Parameters: v_min (minimum air speed requirement) = 0.25 m / s (obtained by checking the standard table); S (roadway section) = 10 m². Calculation: Q_speed = 0.25 * 10 * 60 = 150 m³ / min.

[0094] Example 4: Total air volume decision (decision node): Rule: Q_total = Max(Q_gas, Q_people, Q_speed); Decision: Q_total = Max(750, 200, 150) = 750 m³ / min.

[0095] Data output: Generate a "standard atlas B" that fully complies with the current standard. This atlas represents the ideal calculation logic and results under the same initial conditions.

[0096] Step four: Atlas comparison and difference identification: The system receives "Current Atlas A" (reflecting the original procedure document logic) and "Standard Atlas B" (independently generated based on the specification) as input, starts the atlas comparison engine, and performs structural alignment and semantic comparison on the two directed acyclic graphs. Specifically, the following three dimensions are analyzed for differences: Structural differences: Compare the topological structure integrity of the two atlases. For example, if "Standard Atlas B" contains the [calculate air volume by number of people] node, but "Current Atlas A" completely lacks this node, it is determined that the original document has missed the necessary step of calculating air volume by the number of people, which is a structural omission. Rule differences: Under the premise of node function correspondence, compare whether the calculation logic or reference basis is compliant. For example, "Standard Atlas B" uses the Max() function (takes the maximum value of each calculation air volume) for the total air volume decision node, while "Current Atlas A" incorrectly uses the Average() function; or "Current Atlas A" refers to a safety factor K=1.2 (according to the old standard that has been abolished) when calculating gas air volume, while "Standard Atlas B" uses K=1.5 according to the current specification, which is a rule error. Numerical differences: When the structure and rules are consistent, further compare the numerical results of the nodes. For example, both atlases contain the [calculate air volume by gas] node and have the same calculation formula, but "Current Atlas A" results in 500m³ / min, while "Standard Atlas B" correctly results in 750m³ / min, indicating that the original document may have calculation errors, parameter entry errors, or unit conversion errors, etc. numerical deviations.

[0097] The system automatically classifies, locates, and records the above three types of differences, and finally outputs a clear and traceable difference report, including the location of each difference point (corresponding to the document chapter or sentence), type (structure / rule / numerical), specific description, and compliance suggestions, providing accurate basis for procedure revision and review decisions.

[0098] Step five: generate a traceable review report with links: The system receives the user uploaded engineering regulation document (supports.doc,.docx,.pdf,.wps, etc.), converts the scanned copy into text by using OCR technology, and directly extracts the content of the electronic document; synchronously analyzes the chapter structure, table, formula and key engineering statements, and outputs the structured original text data. Based on the structured text, the named entity recognition (NER) and relationship extraction (RE) technologies are used to automatically extract the engineering parameter nodes (such as [gas emission: 5m³ / min]) and their values, and infer the dependency relationship according to the context logical keywords (such as “according to… calculation” and “take the maximum value”), and construct a directed acyclic graph (DAG) form “status atlas A” to completely restore the data flow and reasoning logic of the original document. The initial parameters (such as roadway section, gas emission, number of workers, etc.) are extracted from the “status atlas A” and input into the independent calculation engine embedded with industry specifications and standard algorithms, and the full-process calculation is strictly performed according to the current technical standards to generate a structure-aligned and logically compliant “standard atlas B”. The “status atlas A” and “standard atlas B” are input into the atlas comparison engine for automatic comparison from three dimensions: structural differences: identify missing or redundant calculation items (such as the lack of “air volume calculation per person” node); rule differences: detect calculation logic or reference standard errors (such as using Average() instead of Max(), or referencing outdated safety factors); numerical differences: compare the calculation results under the premise of consistent structure and rules (such as 500m³ / min vs 750m³ / min). The structured difference list containing all the differences is output.

[0099] The difference list is formatted into an interactive HTML review report, and each difference contains: problem description, original document snapshot, standard basis, and attached hyperlink; clicking the link can display “status atlas A” and “standard atlas B” side by side, highlight the difference nodes and associated paths, and realize instant positioning and root tracing of the problem.

[0100] Step six: output the review report As the final output link of the review mode, the system formally outputs the interactive HTML review report generated in step five for engineers, technical review personnel or management personnel to review. The report presents all the identified structural, rule and numerical differences, and each problem is attached with original document evidence, standard basis and traceable atlas link, supporting quick positioning of the problem root. The user can revise the original regulation document based on the report to ensure that it meets the current specifications and engineering safety requirements, thereby realizing the final goal of intelligent and closed-loop review of the regulation.

[0101] In an embodiment of the present application, as Figure 2The generation mode shown: the generation mode assists engineers to create a brand-new engineering regulation document from scratch through visual interaction, efficiently and accurately. Specifically, it can include the following steps: Step 1: Input initial conditions: the user inputs the initial conditions of the project through the form on the front-end interface. The data content includes but is not limited to “tunnel parameters” such as tunnel length, cross section, “gas data” such as absolute emission, “personnel information” such as maximum number of workers, etc. For example, the engineer inputs “tunnel cross section: 10 m²”, “minimum air speed: 0.25 m / s”, and “absolute gas emission: 5 m³ / min” on the interface. These initial conditions provide basic data for the subsequent construction and calculation of the logic graph.

[0102] Step 2: AI dynamically constructs / updates visual logic graph: the system dynamically constructs and updates a visual logic graph on the front-end interface in real time while the user inputs data. For example, when the user inputs [tunnel cross section: 10 m²], an input node appears on the interface immediately; then input [minimum air speed: 0.25 m / s], another input node appears on the interface. The system discovers that these two parameters can be used to calculate air volume according to the built-in knowledge base, and automatically creates a [calculate air volume according to air speed] calculation node, and points to it from the previous two input nodes with two edges, while calculating and displaying the result on the node in real time: 150 m³ / min. The final output is a visual logic graph that grows in real time with user input.

[0103] Step 3: User interaction and sand table deduction: this step is the core of the whole mode, the user can view details or modify parameters for “What-If” analysis by clicking the node. For example, the user clicks the [calculate air volume according to gas] node on the graph, the system pops up the details window of the node, showing its formula / rule, input value and source, output result, etc. If the engineer temporarily modifies the value of “gas emission” from 5 to 8, the system will mark the node as “user manually overridden” and immediately recalculate the values of the node and its downstream nodes. All affected nodes (such as gas emission, calculate air volume according to gas, total air demand) and their paths will be highlighted on the interface, so that the chain reaction caused by modification is clear at a glance.

[0104] Step 4: One-click generation of standard procedure document: After the engineers confirm the entire logical map is correct through interaction and deduction, they click the "Generate Procedure" button. The system traverses the final confirmed logical map, extracts the final calculation results (such as [Total required air volume: 1200 m³ / min]) and key calculation processes and bases from the key nodes. Then, the system accurately fills the extracted data into the pre-set procedure document template that conforms to national or industry standards. The template contains placeholders, such as "The final determination of the total required air volume is {{Q_total}} m³ / min, mainly based on the calculation result of gas emission." The final output is a filled and generated document object.

[0105] Step 5: Output the final procedure document: As the final product of the generation mode, the system exports the document object generated in step four as a final procedure document (such as.docx or.pdf format) that is format-specification, data-accurate, and logic-clear. This document can be directly used for archiving or publishing, ensuring the standardization and compliance of the procedure. In this way, engineers and technical management personnel can efficiently generate high-quality procedure documents, reducing errors and omissions in manual writing.

[0106] In summary, based on the above technical solutions, the application has the following technical advantages in application: "White-box" generation process solves the trust problem: The application completely transforms the traditional AI "black box" calculation process into a visible, controllable, and auditable "white box". Through the visual logical map, the source and calculation path of each data are clear and traceable, fundamentally solving the engineers' distrust of AI in safety-related fields.

[0107] Create a new human-computer collaboration paradigm and improve expert efficiency: The application provides an efficient collaboration mode of "AI responsible for tedious calculations and connections, and experts focus on key audits and decisions". Experts no longer need to perform repetitive manual calculations, but can focus their efforts on auditing logic, controlling key parameters, and using the "instant recalculation" function for risk deduction, greatly improving decision-making quality and work efficiency.

[0108] Strong binding of data and logic ensures content consistency: This method ensures that each data in the final generated procedure document can find its unique and complete calculation path and rule basis in the logical map. It eliminates the internal data inconsistency problems caused by manual copying and modification in traditional document preparation.

[0109] Sand table deduction capability: The core "instant recalculation" function provides an interactive sand table for engineers. Engineers can freely explore the impact of different parameter changes on the entire system, conduct rapid risk assessment and scheme optimization.

[0110] Fusion of generation and review, covering the whole life cycle of the procedure: the logical graph technology of the present application can not only be used to create new procedures, but also be used to review existing procedures efficiently. Through graph comparison, hidden errors and non-standard places can be quickly located, providing unified and efficient technical support for the whole life cycle management of engineering procedures.

[0111] Embodiment 3 The following will take the generation and review of a mine tunneling working face local ventilation procedure as an example to describe the embodiments of the present application in detail.

[0112] Scenario one: generating a new procedure: Initialization and graph construction: an engineer newly creates a task and inputs the initial conditions: roadway length 800 meters, cross section 16 m², gas emission 5 m³ / min. A visual logical graph appears on the system interface immediately, which initially contains several input nodes such as "roadway length", "cross section", "gas emission". Then, the AI automatically expands the graph and adds parallel calculation nodes such as "calculate air volume according to gas", "calculate air volume according to air speed", "calculate air volume according to number of people", etc., which are all connected from the corresponding input nodes.

[0113] Interactive review: the engineer clicks the "calculate air volume according to gas" node, and a pop-up window shows:

formula

input

result

[0114] Sand table deduction: the engineer considers the uncertainty of geological conditions and wants to evaluate the consequences of a 50% increase in gas emission. He directly modifies 5 m³ / min to 7.5 m³ / min in the pop-up window of the "gas emission" node.

[0115] Instant recalculation and visual feedback: at the moment he confirms the modification, the "calculate air volume according to gas" node, the downstream "total required air volume" decision node, and the more downstream "fan pressure calculation" and "fan selection" nodes on the graph all turn red. The engineer clicks the "fan selection" node and finds that its state has changed from "meet requirements" to "warning: insufficient static pressure". Through this deduction, the engineer predicts the potential risk and decides to select a higher level fan in the scheme to retain a safety margin.

[0116] One-key generation: after completing all reviews and adjustments, the engineer clicks "generate document". The system generates a complete and standard Word version of "XX working face ventilation procedure" in a few seconds, all the data of which are completely consistent with the final confirmed state of the graph.

[0117] Scenario Two: Reviewing Old Regulations 1) Importing and Mapping: An engineer drags a PDF file into the system. The system starts a review mode, parses the document through OCR and information extraction techniques, and builds a "Current Map A" reflecting the old regulation. Meanwhile, according to the initial conditions in the document, it independently calculates and generates a "Standard Map B" based on the latest standards.

[0118] 2) Map Comparison and Report Generation: The map comparison engine starts working and discovers the following typical problems: Case One: Numerical Discrepancy (Numerical Difference): Problem Location: The comparison engine finds that the "wind tunnel air leakage rate" node value in "Current Map A" is 0.2, while in "Standard Map B", according to the latest air leakage rate calculation standard, the calculation result under the same conditions should be 0.25.

[0119] Report Presentation: The system generates a review report, "

Numerical Risk

[0120] Case Two: Missing Calculation Logic (Structural Difference): Problem Location: The comparison engine finds that there is a calculation node called "Calculate Air Volume According to Minimum Wind Speed" in "Standard Map B", which is a necessary calculation step according to the "Coal Mine Safety Regulations". However, this node and its related calculation path are completely missing in "Current Map A".

[0121] Report Presentation: The review report generates a "

Structural Defect

[0122] Case Three: Decision Rule Error (Rule Difference): Problem positioning: The old procedure correctly calculated the "gas-based air volume" (result 120 m³ / min) and "person-based air volume" (result 80 m³ / min). However, when determining the "air distribution volume", the value in the document is 100 m³ / min. The comparison engine found that the rule of the "air distribution volume" decision node in "Current Atlas A" was incorrectly identified as "taking the average", while the corresponding node in "Standard Atlas B" has a clear rule of "taking the maximum value among all calculated air volumes".

[0123] Report presentation: The report contains the entry: "[Rule Error] 'Air Distribution Volume' decision logic error. Current value is 100 m³ / min (suspected to take the average), which is inconsistent with the standard requirement of 'taking the maximum value' (should be 120 m³ / min). Risk analysis: causes the air distribution volume to be lower than the required air volume for gas emission, which poses a safety hazard." Clicking on it, the system highlights the "air distribution volume" decision node and pops up a window comparing the decision rules and input / output data of the two atlases at this node.

[0124] Embodiment 4 The embodiments of the present application also provide an engineering regulation generation and verification system 10 based on an interactive logic atlas, as shown in the figure, the system 10 comprises: Figure 3 a document analysis and atlas generation module 100 for analyzing the to-be-inspected engineering regulation document and generating a current atlas A reflecting the calculation logic thereof, and independently constructing a standard atlas B based on the initial conditions and the latest industry standards extracted from the document; a multi-dimensional comparison and analysis module 200 for performing structural, rule and numerical difference analysis on the current atlas A and the standard atlas B through a multi-dimensional comparison algorithm, identifying calculation path missing, logic rule error or parameter calculation deviation; an interactive review report generation module 300 for generating an interactive review report containing a list of differences, wherein each difference is associated with a corresponding node of the current atlas A and the standard atlas B, and the problem tracing is immediately located through a hyperlink; a parameter modification and dependent recalculation module 400 for allowing a user to modify parameters or rules of the current atlas A in the review report, triggering immediate recalculation of the dependent relationship chain and updating the atlas state to verify whether the corrected calculation logic meets the constraint conditions of the standard atlas B.

[0125] Further, the document analysis and atlas generation module is further used for: S11, adopting an optical character recognition (OCR) technology to process a scanned PDF or a picture format document, converting it into an editable text stream, and extracting engineering parameter nodes including a roadway cross section, a gas emission amount and a number of workers through a named entity recognition (NER) model in a natural language processing technology. ​S12, based on the context information in the document, a key phrase is identified by using a relation extraction (RE) model, a directed edge between parameter nodes is constructed, and a directed acyclic graph (DAG) structure reflecting the calculation logic of the original document is formed.

[0126] Further, the multi-dimensional comparative analysis module is further used for: Topological alignment is performed on the structural differences, missing nodes are detected by traversing the node set of the status atlas A and the standard atlas B, and corresponding problem descriptions are generated.

[0127] Further, the interactive review report generation module is further used for: Each difference point is bound to the corresponding node of the status atlas A and the standard atlas B, and the problem traceability is instantaneously positioned through the hyperlinks in HTML format, which displays the complete calculation path and parameter source of the difference node in the two atlases after clicking.

[0128] Further, the parameter modification and dependent recalculation module is further used for: When the user modifies the parameter, the system marks the node as user manual override, and triggers the instant recalculation of the downstream node based on the dependency relationship chain, and the recalculation process includes:

[0129]

[0130]

[0131]

[0132] Wherein, Q is the gas emission amount, Q is the gas unevenness coefficient, Q is the allowable gas concentration, Q is the number of workers, Q is the air volume per person, Q is the minimum air speed, Q is the roadway cross section.

[0133] The engineering regulation generation and verification system based on the interactive logic atlas of the embodiment of the application, by version checking the difference points in the review report and the latest industry standard in the knowledge base, and automatically generating version risk prompts and clause comparison views when detecting outdated standard references, further improves the standardization and timeliness of regulation review, effectively avoids technical compliance risks caused by lagging standard versions.

[0134] The above merely describes preferred embodiments of the present application, and is not intended to limit the present application. The present application can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of the present application.

[0135] In the description of the present specification, the description referring to the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. The illustrative representation of the above terms in the present specification does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples, without contradiction.

[0136] In addition, the terms "first", "second" are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

Claims

1. A method for generating and verifying engineering specifications based on interactive logic graphs, characterized in that, include: S1, parse the engineering specification document to be reviewed and generate a current status map A that reflects its calculation logic, and independently construct a standard map B based on the initial conditions extracted from the document and the latest industry standards; S2 uses a multi-dimensional comparison algorithm to analyze the structural, rule, and numerical differences between the current map A and the standard map B, and identifies missing calculation paths, logical rule errors, or parameter calculation deviations. S3 generates an interactive review report containing a list of differences, where each difference is associated with a corresponding node in the current status map A and the standard map B, and the source of the problem can be located instantly through hyperlinks. S4 allows users to modify the parameters or rules of the current graph A in the review report, triggering an immediate recalculation of the dependency chain and updating the graph status to verify whether the modified calculation logic meets the constraints of the standard graph B.

2. The method as described in claim 1, characterized in that, The process of S1, which involves parsing the engineering specification document to be reviewed and generating a current status map A reflecting its calculation logic, also includes: S11 uses Optical Character Recognition (OCR) technology to process scanned PDF or image documents, converting them into editable text streams, and extracts engineering parameter nodes through Named Entity Recognition (NER) model in Natural Language Processing (NLP). The engineering parameter nodes include tunnel cross-section, gas emission rate, and number of workers. S12, based on the contextual information in the document, uses the relation extraction (RE) model to identify key phrases, constructs directed edges between parameter nodes, and forms a directed acyclic graph (DAG) structure that reflects the computational logic of the original document.

3. The method as described in claim 1, characterized in that, The S2 method, which uses a multi-dimensional comparison algorithm to analyze the structural, regular, and numerical differences between the current map A and the standard map B, also includes: S21. Perform topological alignment on structural differences by traversing the node sets of the current graph A and the standard graph B, detecting missing nodes and generating corresponding problem descriptions.

4. The method as described in claim 1, characterized in that, The generation of an interactive review report containing a list of differences in S3 also includes: S31, bind each difference point to the corresponding node of the current status map A and the standard map B, and realize the instant location of the problem source through the HTML hyperlink. After clicking the hyperlink, the complete calculation path and parameter source of the difference node in the two maps are displayed side by side.

5. The method as described in claim 1, characterized in that, The S4, which allows users to modify parameters or rules of the current status map A in the review report, also includes: S41, When a user modifies parameters, the system marks the node as manually overwritten by the user and triggers immediate recalculation of downstream nodes based on the dependency chain. The recalculation process includes: in, For gas emission volume, The gas unevenness coefficient, To allow for gas concentration, For the number of workers, per capita air volume required At the lowest wind speed, This is the cross-section of the tunnel.

6. A system for generating and verifying engineering specifications based on an interactive logic graph, characterized in that, include: The document parsing and graph generation module is used to parse the engineering specification documents to be reviewed and generate a current status graph A that reflects its calculation logic. At the same time, it independently constructs a standard graph B based on the initial conditions extracted from the document and the latest industry standards. The multi-dimensional comparison and analysis module is used to analyze the structural, rule and numerical differences between the current map A and the standard map B through a multi-dimensional comparison algorithm, and to identify missing calculation paths, logical rule errors or parameter calculation deviations. The interactive review report generation module is used to generate an interactive review report containing a list of differences. Each difference is associated with a corresponding node in the current status map A and the standard map B, and the problem can be located instantly through hyperlinks. The parameter modification and dependency recalculation module allows users to modify the parameters or rules of the current graph A in the review report, triggering the immediate recalculation of the dependency chain and updating the graph status to verify whether the modified calculation logic meets the constraints of the standard graph B.

7. The system as described in claim 6, characterized in that, The document parsing and graph generation module is also used for: S11 uses Optical Character Recognition (OCR) technology to process scanned PDF or image documents, converting them into editable text streams, and extracts engineering parameter nodes through Named Entity Recognition (NER) model in Natural Language Processing (NLP). The engineering parameter nodes include tunnel cross-section, gas emission rate, and number of workers. S12, based on the contextual information in the document, uses the relation extraction (RE) model to identify key phrases, constructs directed edges between parameter nodes, and forms a directed acyclic graph (DAG) structure that reflects the computational logic of the original document.

8. The system as described in claim 6, characterized in that, The multi-dimensional comparison analysis module is also used for: Topological alignment is performed on structural differences by traversing the node sets of the current graph A and the standard graph B, detecting missing nodes and generating corresponding problem descriptions.

9. The system as described in claim 6, characterized in that, The interactive review report generation module is also used for: Each point of difference is bound to the corresponding node in the current map A and the standard map B. The problem source is located instantly through the hyperlink in HTML format. After clicking the hyperlink, the complete calculation path and parameter source of the point of difference in the two maps are displayed side by side.

10. The system as described in claim 6, characterized in that, The parameter modification and dependency recalculation module is also used for: When a user modifies parameters, the system marks the node as manually overwritten by the user and triggers an immediate recalculation of downstream nodes based on the dependency chain. The recalculation process includes: in, For gas emission volume, The gas unevenness coefficient, To allow for gas concentration, For the number of workers, per capita air volume required At the lowest wind speed, This is the cross-section of the tunnel.

Images