Coal mine regulation intelligent matching and structured interpretation method and device, and storage medium

Abstract

This invention relates to the field of intelligent processing technology for coal mine safety regulations, and in particular to a method, apparatus, equipment, and computer storage medium for intelligent matching and structured interpretation of coal mine regulations. This invention acquires coal mine regulations text and performs hierarchical decomposition and metadata extraction. It then uses a pre-trained language model to generate regulation vectors to construct a regulatory knowledge base. Simultaneously, it acquires coal mine production scenario text and uses the same encoding model to generate production scenario vectors. The semantic similarity between the two is calculated to filter a set of candidate regulatory clauses. This achieves automatic semantic matching between coal mine production scenarios and regulatory clauses, overcoming the shortcomings of traditional keyword matching in capturing implicit semantics and complex logical relationships, and significantly improving the coverage, accuracy, and intelligence level of regulation retrieval.

Description

Technical Field

[0001] This invention relates to the field of intelligent processing technology of coal mine safety regulations, and in particular to a method, apparatus, equipment and computer storage medium for intelligent matching and structured interpretation of coal mine regulations. Background Technology

[0002] Coal mine safety production relies on a multi-layered constraint system of national regulations and industry standards. With the advancement of intelligent mine construction, underground monitoring and equipment management systems have formed a multi-source heterogeneous data ecosystem. However, existing methods for applying regulations mainly rely on keyword matching or manual experience-based judgment, failing to establish a deep semantic association mechanism between production scenarios and regulatory clauses. Due to the complex conditional logic and implicit constraints of regulatory texts, traditional technologies struggle to automatically decompose and structure unstructured clauses, resulting in static regulations failing to effectively link with real-time dynamic data. The resulting matching omissions, misjudgments, or delayed responses keep safety management at the post-event verification stage, severely hindering the improvement of proactive risk prevention and intelligent decision-making capabilities in coal mines. Summary of the Invention

[0003] Therefore, the technical problem to be solved by the present invention is to overcome the problems in the prior art that rely on manual retrieval and subjective interpretation, resulting in low efficiency and easy omissions in regulatory matching, and that unstructured regulatory texts are difficult to convert into computable rules, thus making it impossible to link with real-time production data.

[0004] To address the aforementioned technical problems, this invention provides a method for intelligent matching and structured interpretation of coal mine regulations, comprising:

[0005] The text of coal mine regulations is obtained and hierarchically split and metadata is extracted. A pre-trained language model is used to semantically encode the split regulations to generate regulations vectors, and a regulations knowledge base containing regulations vectors is constructed. The text of a coal mine production scenario is obtained and standardized. A production scenario vector is generated using the same encoding model as the pre-trained language model. The semantic similarity between the production scenario vector and each legal vector in the legal knowledge base is calculated. A set of candidate legal clauses whose similarity meets the preset conditions is selected. The natural language parsing tool is invoked to perform structured parsing on each candidate regulation clause in the candidate regulation clause set, extracting information on the work object, constraints, safety parameter thresholds and handling measures, and generating structured safety rules containing logical judgment expressions; The real-time collected coal mine production data is substituted into the logical judgment expression in the structured safety rules to determine compliance. Based on the judgment result, a risk level assessment and rectification suggestions are generated and output.

[0006] Preferably, the step of acquiring coal mine regulations text and performing hierarchical decomposition and metadata extraction, using a pre-trained language model to semantically encode the decomposed regulations into regulations vectors, and constructing a regulations knowledge base containing the regulations vectors includes: Coal mine safety-related regulatory texts are obtained from a multi-source regulatory platform, and standardized regulatory texts are obtained by unifying the encoding and formatting. The standardized regulatory text is split into chapters, clauses, and sentences, and the clause number, clause title, applicable objects, and regulatory level meta-information are extracted to obtain split clauses with meta-information. A pre-trained Chinese legal language model is invoked to semantically encode the clause text in the split clauses containing meta-information, generating a legal clause vector. The vectors of the aforementioned legal provisions are stored in a vector database to construct a legal knowledge base.

[0007] Preferably, the steps of acquiring coal mine production scene text and performing standardization processing, generating production scene vectors using the same encoding model as the pre-trained language model, calculating the semantic similarity between the production scene vectors and each regulation vector in the regulatory knowledge base, and selecting a set of candidate regulatory clauses whose similarity meets preset conditions include: Original production scene descriptions are obtained from multi-source data terminals at the coal mine production site, and noise reduction, word segmentation and standardization are performed to obtain a standardized production scene text set. The pre-trained language model is invoked to encode each production scenario description in the standardized production scenario text set, generating a production scenario vector. Normalization and dimensionality reduction are performed on the production scene vector to obtain the normalized and dimensionality-reduced scene vector; Calculate the cosine similarity between the normalized and dimensionality-reduced scene vector and each regulation vector in the regulation knowledge base, and select regulation clauses with similarity greater than or equal to the matching similarity threshold as a set of candidate regulation clauses.

[0008] Preferably, the step of calling a natural language parsing tool to perform structured parsing on each candidate regulatory clause in the candidate regulatory clause set, extracting information on the work object, constraints, safety parameter thresholds, and handling measures, and generating structured safety rules containing logical judgment expressions includes: The AI ​​agent invokes natural language processing tools to perform structured analysis, entity recognition, and semantic dependency parsing on candidate regulatory clauses, automatically extracting key information such as job type, constraints, safety thresholds, and handling measures. A rule mapping function is established based on the extracted key information to map candidate legal clauses into a structured field vector; For clauses containing complex combinations of conditions, a logical judgment expression is constructed using Boolean logic formulas. This logical judgment expression combines multiple condition judgment units through logical AND operations to determine whether the production status is compliant.

[0009] Preferably, the step of substituting real-time collected coal mine production data into the logical judgment expression in the structured safety rules for compliance determination, and generating and outputting risk level assessments and rectification suggestions based on the determination results includes: Acquire real-time coal mine production data and construct a real-time production data vector, then substitute the real-time production data vector into the logical judgment expression in the structured safety rules to calculate the truth value; Compliance indicators are generated based on the calculation results using indicator functions. These compliance indicators are used to indicate whether the production status complies with regulatory requirements or whether there are any violations. Based on the aforementioned compliance indicators and the degree of deviation between the real-time production data vector and the regulatory constraint thresholds, the risk level is calculated using a risk level assessment function, and corresponding rectification suggestions are generated based on the risk level.

[0010] Preferably, the calculation of the risk level based on the deviation between the compliance indicators and the real-time production data vector and the regulatory constraint threshold, using a risk level assessment function, includes: Extract key production parameters and corresponding regulatory constraint thresholds from real-time production data vectors; When multiple key production parameters simultaneously exceed the corresponding regulatory thresholds, the risk level is determined to be high. If any one of the key production parameters exceeds the corresponding regulatory threshold, it is classified as medium risk. When all key production parameters are less than or equal to the corresponding regulatory thresholds, the risk level is determined to be low.

[0011] Preferably, the method further includes: The matched regulation number, structured interpretation results, compliance status, and rectification suggestions are pushed to the coal mine safety management platform to achieve closed-loop linkage between regulatory provisions, production data, and safety management.

[0012] This invention also provides a device for intelligent matching and structured interpretation of coal mine regulations, comprising: The regulations acquisition module is used to acquire coal mine regulations texts and perform hierarchical splitting and metadata extraction. The semantic modeling module is used to semantically encode the split legal clauses using a pre-trained language model to generate legal vectors and build a legal knowledge base containing legal vectors. The vector matching module is used to acquire coal mine production scene text and perform standardization processing, generate production scene vectors using the same encoding model as the pre-trained language model, calculate the semantic similarity between the production scene vectors and each legal vector in the legal knowledge base, and filter out a set of candidate legal clauses whose similarity meets the preset conditions. The structured interpretation module is used to call natural language parsing tools to perform structured parsing on each candidate legal clause in the candidate legal clause set, extract information on the work object, constraints, safety parameter thresholds and disposal measures, and generate structured safety rules containing logical judgment expressions; The compliance determination module is used to substitute real-time collected coal mine production data into the logical judgment expression in the structured safety rules to determine compliance, and generate and output risk level assessment and rectification suggestions based on the determination results.

[0013] This invention also provides a device for intelligent matching and structured interpretation of coal mine regulations, comprising: Memory, used to store computer programs; A processor is used to implement the steps of the above-mentioned intelligent matching and structured interpretation method for coal mine regulations when executing the computer program.

[0014] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described intelligent matching and structured interpretation method for coal mine regulations.

[0015] The technical solution of the present invention has the following advantages over the prior art: The intelligent matching and structured interpretation method for coal mine regulations described in this invention acquires coal mine regulations text, performs hierarchical decomposition and metadata extraction, generates regulation vectors using a pre-trained language model to construct a regulation knowledge base, and simultaneously acquires coal mine production scenario text and generates production scenario vectors using the same encoding model. The semantic similarity between the two is calculated to filter a set of candidate regulation clauses, thereby achieving automatic semantic matching between coal mine production scenarios and regulation clauses. This overcomes the shortcomings of traditional keyword matching, which cannot capture implicit semantics and complex logical relationships, significantly improving the coverage, accuracy, and intelligence level of regulation retrieval. Furthermore, this invention calls natural language parsing tools to perform structured parsing of candidate regulation clauses, extracting information on work objects, constraints, safety parameter thresholds, and handling measures, generating structured safety rules containing logical judgment expressions. This transforms unstructured regulation text into computer-executable structured rules, solving the problems of low efficiency and error-proneness in manual interpretation, realizing the engineering expression of regulatory knowledge, and making regulatory rules computable. Finally, this invention substitutes real-time collected coal mine production data into the logical judgment expression in structured safety rules for compliance determination. Based on the determination result, it generates and outputs risk level assessment and rectification suggestions, realizing automated compliance determination and proactive risk warning based on real-time production data. This effectively avoids the subjectivity and lag of manual judgment, elevates safety management from post-verification to proactive prevention and control, and significantly improves the intelligence level, response speed, and decision-making ability of coal mine safety management. Attached Figure Description

[0016] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the implementation of a method for intelligent matching and structured interpretation of coal mine regulations provided by this invention. Figure 2 This is a structural block diagram of a coal mine regulation intelligent matching and structured interpretation device provided in an embodiment of the present invention. Detailed Implementation

[0017] The core of this invention is to provide a method, device, equipment, and computer storage medium for intelligent matching and structured interpretation of coal mine regulations. This method can effectively realize automatic semantic matching between production scenarios and regulatory clauses, as well as the conversion of unstructured regulatory texts into structured safety rules, supporting real-time compliance judgment and proactive risk warning.

[0018] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please refer to Figure 1. Figure 1 The flowchart illustrates the implementation of a method for intelligent matching and structured interpretation of coal mine regulations provided by this invention; the specific operation steps are as follows: S101: Obtain the text of coal mine regulations and perform hierarchical decomposition and metadata extraction. Use a pre-trained language model to semantically encode the decomposed regulations to generate regulations vectors and construct a regulations knowledge base containing regulations vectors. S102: Obtain the text of the coal mine production scene and perform standardization processing; generate the production scene vector using the same encoding model as the pre-trained language model; calculate the semantic similarity between the production scene vector and each legal vector in the legal knowledge base; and select a set of candidate legal clauses whose similarity meets the preset conditions. S103: Call a natural language parsing tool to perform structured parsing on each candidate regulation clause in the candidate regulation clause set, extract information on the work object, constraints, safety parameter thresholds and handling measures, and generate structured safety rules containing logical judgment expressions; S104: Substitute the real-time collected coal mine production data into the logical judgment expression in the structured safety rules to determine compliance, and generate and output risk level assessment and rectification suggestions based on the judgment results.

[0020] In some embodiments, the specific implementation of constructing a regulatory knowledge base containing regulatory vectors as described in S101 is as follows. First, the input sources of the system cover national regulatory databases, industry standard platforms, and enterprise internal safety management systems. The collected objects are national standards, industry norms, local regulations, and enterprise procedure texts related to coal mine safety. The processing action is to perform UTF-8 encoding and format standardization on the collected multi-source regulatory texts to eliminate character set differences and typesetting noise, and the output result is standardized regulatory text. Subsequently, using the standardized regulatory text as input, it is recursively split into three levels: chapter, clause, and sentence, while extracting key meta-information such as clause number, clause title, applicable object, and regulatory level, and the output result is split clauses with meta-information. Next, the pre-trained Chinese legal language model Legal BERT in the natural language processing toolchain is called to perform semantic encoding on the clause text in the split clauses with meta-information. Specifically, the text of the i-th regulatory clause is selected. As input data, the Legal BERT model is invoked to perform semantic feature extraction, generating an initial semantic vector. Based on this, the norm of the initial semantic vector is calculated, and the cosine similarity formula is applied to calculate the semantic similarity between the text of the i-th legal clause and other legal clause texts, resulting in a semantic similarity matrix. Its calculation formula is ,in and These represent the vector representations of different clauses. Representing vectors The norm of the semantic similarity matrix is ​​used to optimize and adjust the position of the initial semantic vector in the vector space, ultimately generating the vector representation of the i-th legal clause. Satisfying the relation ,in This indicates that semantic encoding processing is performed by calling a pre-trained Chinese legal language model. This represents the text of the i-th regulatory clause. Finally, the generated regulatory clause vector is stored in the vector database Milvus, completing the construction of the regulatory knowledge base.

[0021] In other embodiments, when optimizing the spatial position of the initial semantic vector using a cosine similarity matrix during the generation of regulatory clause vectors, the similarity matrix can also be weighted and adjusted in conjunction with the meta-information of the clauses (such as regulatory level and applicable objects) to enhance the expression of the hierarchical relationship of regulations.

[0022] In one specific embodiment: the text of the "Coal Mine Safety Regulations" is obtained from the national legal database, broken down by chapter, section, and article, and metadata such as "Article X", "gas concentration", and "mining face" are extracted. The Legal BERT model is then used to generate vector representations, which are stored in the Milvus vector database to construct a legal knowledge base.

[0023] Specifically: the pre-trained Chinese legal language model Legal BERT is a BERT model fine-tuned based on legal domain corpora, enabling it to better understand the professional terminology and logical structure in legal texts. The cosine similarity... The value range is [-1, 1], and the closer the value is to 1, the more semantically similar the two legal clauses are. The final generated legal clause vector... The dimension is 1024.

[0024] It should be noted that this step transforms coal mine regulations from static documents to dynamic semantic vectors, overcoming the shortcomings of traditional keyword matching in capturing implicit semantics and complex logical relationships. This lays a data foundation for the accurate semantic association between production scenarios and regulatory clauses, and significantly improves the coverage and intelligence level of regulatory retrieval.

[0025] Based on the above embodiments, in some embodiments, the step of acquiring coal mine regulations text and performing hierarchical splitting and metadata extraction, using a pre-trained language model to semantically encode the split regulations clauses to generate regulations vectors, and constructing a regulations knowledge base containing regulations vectors includes: Coal mine safety-related regulatory texts are obtained from a multi-source regulatory platform, and standardized regulatory texts are obtained by unifying the encoding and formatting. The standardized regulatory text is split into chapters, clauses, and sentences, and the clause number, clause title, applicable objects, and regulatory level meta-information are extracted to obtain split clauses with meta-information. A pre-trained Chinese legal language model is invoked to semantically encode the clause text in the split clauses containing meta-information, generating a legal clause vector. The vectors of the aforementioned legal provisions are stored in a vector database to construct a legal knowledge base.

[0026] In some embodiments, when splitting regulatory text by level, the clause text can be further split by sentence level, splitting long clauses into multiple independent sentence units, and maintaining the mapping relationship between clause numbers and sentences to improve the fine granularity of subsequent semantic encoding.

[0027] In other embodiments, principal component analysis (PCA) can be performed on the generated regulatory clause vectors before storing them in the vector database to compress the high-dimensional vectors to a lower dimension, thereby improving retrieval efficiency.

[0028] In one specific embodiment: taking Article 138 of the "Coal Mine Safety Regulations" as an example, which states that "the gas concentration in the airflow of the mining face shall not exceed 1.0%", after splitting the article by clause, the clause number "Article 138", the applicable object "mining face", and the regulatory level "national standard" are extracted. Then, Legal BERT is used to generate a vector representation. Stored in the Milvus database.

[0029] Specifically: the semantic encoding process is achieved through the formula Implementation, in which This represents the text of the i-th legal provision. This indicates that semantic encoding processing is performed by calling a pre-trained Chinese legal language model. This represents the vector representation of the i-th legal provision.

[0030] It should be noted that this embodiment converts the regulatory text into a vector representation, making semantically similar clauses closer together in the vector space, thereby supporting efficient subsequent retrieval and association analysis.

[0031] Based on the above embodiments, in some embodiments, the steps of acquiring coal mine production scene text and performing standardization processing, generating production scene vectors using the same encoding model as the pre-trained language model, calculating the semantic similarity between the production scene vectors and each regulation vector in the regulatory knowledge base, and selecting a set of candidate regulatory clauses whose similarity meets preset conditions include: Original production scene descriptions are obtained from multi-source data terminals at the coal mine production site. Noise reduction, word segmentation, and standardization processes are then performed to obtain a standardized production scene text set, denoted as […]. Where SS represents a standardized production scenario text set, This represents the descriptions of production scenarios from the 1st to the mth. Invoke the same pre-trained language model as in the regulatory coding phase. Encode each production scenario description in the standardized production scenario text set to generate a production scenario vector. Its mathematical expression is ,in This represents the vector representation of the i-th production scenario description. This represents a pre-trained language model. This represents the i-th production scenario description, where d represents the vector dimension; The production scene vector is normalized and dimensionality reduced to obtain the normalized and dimensionality-reduced scene vector. The processing procedure is represented as follows:

[0032] in This indicates the dimensionality reduction operation in principal component analysis. Representing vectors L2 norm, This represents the scene vector after normalization and dimensionality reduction; Calculate the cosine similarity between the normalized and dimensionality-reduced scene vector and each regulation vector in the regulation knowledge base, and select regulation clauses with similarity greater than or equal to the matching similarity threshold as a set of candidate regulation clauses.

[0033] In some embodiments, the multi-source data terminal includes: environmental and gas concentration sensor data from the downhole monitoring system, PLC data from the equipment operation log, work report text, and descriptive information from a manual input terminal.

[0034] In other embodiments, the cosine similarity is calculated using the following formula: ,in This represents the i-th production scene vector after normalization and dimensionality reduction. This represents the vector of the j-th legal clause in the legal knowledge base. Describes the L2 norm of a vector. This represents the calculated semantic similarity. The expression for the candidate legal clause set is: ,in To match the similarity threshold, This indicates Article j of the regulations. Let represent the set of candidate regulatory clauses corresponding to the i-th production scenario.

[0035] In one specific embodiment: the gas concentration at the current mining face is 0.8% obtained from the underground gas sensor. After noise reduction and word segmentation, the standardized text "Gas concentration at the mining face 0.8%" is generated as... Call the same Legal BERT model as in step S101. Generate production scene vectors After normalization and PCA dimensionality reduction, we obtain , and vectors in the legal knowledge base Calculate cosine similarity A threshold τ = 0.75 is set, and legal clauses with a similarity greater than or equal to 0.75 are selected as the candidate set. .

[0036] Specifically: the production scenario vector The generating formula is Where d=1024. The normalized and dimensionality-reduced scene vector... pass The cosine similarity sim has a range of [-1, 1], and the closer the value is to 1, the more semantically relevant the production scenario and the regulatory clauses are.

[0037] It should be noted that this embodiment effectively reduces the interference of high-dimensional data noise on matching accuracy by standardizing the vector representation of the production scenario text and performing PCA dimensionality reduction. At the same time, combined with a strict cosine similarity threshold screening mechanism, it ensures that the candidate regulatory clauses are highly relevant to the current production scenario at the semantic level, which significantly improves the accuracy and real-time performance of regulatory matching.

[0038] Based on the above embodiments, in some embodiments, the step of calling a natural language parsing tool to perform structured parsing on each candidate legal clause in the candidate legal clause set, extracting information on the work object, constraints, safety parameter thresholds, and handling measures, and generating structured safety rules containing logical judgment expressions includes: The AI ​​agent invokes natural language processing tools to perform structured analysis, entity recognition, and semantic dependency parsing on candidate regulatory clauses, automatically extracting key information such as job type, constraints, safety thresholds, and handling measures. Based on the extracted key information, a rule mapping function g is established to map candidate legal clauses. Mapped to a structured field vector ,in and , This represents a structured field vector, where g represents a natural language parsing and logical expression generation function. This indicates the j-th candidate regulatory provision. to Standardized fields representing extracted job types, constraints, safety thresholds, and handling measures; For clauses containing complex combinations of conditions, a logical judgment expression is constructed using Boolean logic formulas. This expression combines multiple condition judgment units through logical AND operations to determine whether the production status is compliant. For clauses containing complex combinations of conditions, identify multiple production scenario parameters within the clause. and their corresponding clause constraint thresholds And determine the relational operators that connect the parameters. The relational operators include greater than (>), less than (<), equal to (=), or belonging to (∈); the production scenario parameters, relational operators, and clause constraint thresholds are substituted into the Boolean logic formula. In this process, multiple conditional judgment units are combined using logical "AND" operations to generate a logical judgment expression used to determine whether the production status is compliant. ,in Terms and Conditions Logical judgment expression, This represents the parameter of the l-th production scenario. This represents the l-th relational operator. This represents the threshold value for the l-th clause constraint. This represents a logical AND operation, where k represents the total number of condition judgment units; the logical judgment expression is... Mapped into a pre-defined structured regulatory template to form executable structured security rules.

[0039] In some embodiments, the natural language parsing tool includes a SpaCy or HanLP engine that performs word segmentation, part-of-speech tagging, named entity recognition, and semantic dependency parsing.

[0040] In other embodiments, the preset structured regulatory template includes job type, restrictions, safety indicators, handling suggestions, and clause numbers, and supports multi-level nesting of chapters, clauses, and sub-clauses. The structured data is stored in a regulatory database in JSON or XML format.

[0041] In one specific embodiment: for candidate regulatory provisions The AI ​​agent uses the SpaCy tool to perform entity recognition, specifying the work type as "mining face," the constraint as "must not exceed," the safety threshold as "1.0%," and the disposal measures as empty. A structured field vector is then generated using the rule mapping function g. ={ ={Operation Type = "Mining Face", Constraints = "≤", Safety Threshold = 1.0, Handling Measures = "None"}}. For the complex clause "If Operation Type = Underground Tunneling and Gas Concentration > 1%, then trigger a safety warning", the production scenario parameters are identified. =Assignment type, =Gas concentration, corresponding constraint threshold =“Underground tunneling” =1%, relational operators = "=", =">, Substitute into the Boolean logic formula This generates logical judgment expressions and maps them into structured regulatory templates to form executable structured security rules.

[0042] Specifically: the logical judgment expression In the middle, when all condition judgment units When both are true, The calculation result is True, indicating that the production status meets the constraints of this clause; if any condition judgment unit is not true, then... The calculation result is False, indicating a violation of the clause.

[0043] It should be noted that this embodiment utilizes an AI agent combined with professional natural language processing tools to automate the conversion of coal mine regulations from unstructured text to structured logical expressions, effectively solving the problems of low efficiency and error-proneness in manual interpretation. The judgment expressions constructed using Boolean logic formulas can accurately describe the combined constraint relationships under complex working conditions, making the regulations computable and providing a precise logical foundation for the automated compliance determination of subsequent real-time production data.

[0044] Based on the above embodiments, in some embodiments, the step of substituting real-time collected coal mine production data into the logical judgment expression in the structured safety rules for compliance determination, and generating and outputting risk level assessments and rectification suggestions based on the determination results includes: Acquire real-time coal mine production data and construct a real-time production data vector. ,in This represents the real-time production data vector collected in the i-th iteration. to Indicates specific parameters such as gas concentration, temperature, or type of operation; Substitute the real-time production data vector into the logical judgment expression in the structured security rule. Perform truth calculation; Compliance indicators are generated based on the calculation results using the indicator function 1{}. ,in, , This represents a compliance indicator, where {1} is an indicator function that takes a value of 1 when the condition within the parentheses is true, and 0 otherwise. This indicates that a real-time data vector will be generated. Substitute into the logical judgment expression The truth value calculation result after; when This indicates that the production status complies with regulatory requirements. The time indicates a violation or exceeding the threshold. The compliance indicator is used to indicate whether the production status complies with regulatory requirements or whether there is a violation. Based on the aforementioned compliance indicators and the degree of deviation between the real-time production data vector and the regulatory constraint thresholds, the risk level is calculated using the risk level assessment function h. Based on the risk level, corresponding rectification suggestions are generated, and the matched regulation number, structured interpretation results, compliance status, and rectification suggestions are pushed to the coal mine safety management platform. In some embodiments, the calculation process of the risk level assessment function h includes: extracting real-time production data vectors. Key production parameters and and the corresponding regulatory thresholds and If satisfied The conditions will determine the risk level. Determined to be high risk; if the following conditions are met The conditions will determine the risk level. Determined as medium risk; if the following conditions are met The conditions will determine the risk level. Determined as low risk, the final risk level is output. .in and This represents two key production parameters in the real-time production data vector. and This represents the corresponding regulatory constraint threshold; ∧ represents the logical "AND" and ∨ represents the logical "OR". Let represent the risk level output by the i-th judgment, and h represent the risk level assessment function.

[0045] In one specific embodiment: A real-time gas concentration of 1.2% is collected, the operation type is "tunneling," and a real-time production data vector is constructed. Substitute into the logical judgment expression generated in step S103 : Operation type = tunneling and gas concentration > 1.0%, true value is calculated as True, compliance indicator This indicates a violation. Furthermore, the gas concentration... =1.2% corresponds to the constraint threshold =1.0%, and simultaneously satisfy > If the risk level is determined to be medium risk (or high risk if multiple key parameters exceed the threshold simultaneously), the system generates a rectification suggestion: "Strengthen ventilation and closely monitor gas concentration," and pushes the regulation number, structured interpretation results, compliance status, and rectification suggestion to the coal mine safety management platform.

[0046] Specifically: the compliance indicators The calculation formula is , where {1} is an indicator function. The complete expression for the risk level assessment function h is: The logic for determining high-risk, medium-risk, and low-risk can be uniformly expressed as:

[0047] It should be noted that this step achieves millisecond-level determination of compliance status by automating the comparison between real-time production data and structured logic expressions, and accurately quantifies risk levels using multi-level threshold logic, effectively avoiding the subjectivity and lag of manual judgment, and significantly improving the proactive prevention and control capabilities of coal mine safety production.

[0048] Based on the above embodiments, in some embodiments, the calculation of the risk level using a risk level assessment function based on the degree of deviation between the compliance indicators and real-time production data vectors and regulatory constraint thresholds includes: Extract key production parameters and corresponding regulatory constraint thresholds from real-time production data vectors; When multiple key production parameters simultaneously exceed the corresponding regulatory thresholds, the risk level is determined to be high. When any one of the key production parameters exceeds the corresponding regulatory threshold, it is classified as medium risk; When all key production parameters are less than or equal to the corresponding regulatory thresholds, the risk level is determined to be low.

[0049] In some embodiments, the key production parameters include: downhole environmental parameters such as gas concentration, temperature, dust concentration, and wind speed, as well as equipment operating parameters such as current, voltage, and rotational speed.

[0050] In other embodiments, the following rectification suggestions are provided: when the risk level is high, the generated rectification suggestions include immediately stopping the operation, starting the ventilation system, and organizing the evacuation of personnel; when the risk level is medium, the rectification suggestions include strengthening monitoring, adjusting operating parameters, and increasing ventilation volume; when the risk level is low, the rectification suggestions include continuous monitoring and regular inspections.

[0051] In one specific embodiment: real-time production data vector medium gas concentration =1.2%, its constraint threshold =1.0%; dust concentration Its constraint threshold Since 1.2% > 1.0% and 10 > 8, and both exceed the constraint threshold, it is judged as high risk, and the rectification suggestion is generated: "Immediately stop mining operations, increase ventilation, and organize personnel evacuation."

[0052] Specifically: the high-risk determination criteria are as follows: The criteria for determining medium risk are: The criteria for determining low risk are: ,in and This represents two key production parameters in the real-time production data vector. and The symbol represents the corresponding regulatory constraint threshold, ∧ represents the logical "AND", and ∨ represents the logical "OR".

[0053] It should be noted that this embodiment uses multi-level threshold logic to accurately quantify the risk level, making the risk assessment results more in line with the actual needs of coal mine safety production, and making it easier for managers to take targeted measures based on the risk level.

[0054] Based on the above embodiments, in some embodiments, the method further includes: The matched regulation number, structured interpretation results, compliance status, and rectification suggestions are pushed to the coal mine safety management platform to achieve closed-loop linkage between regulatory provisions, production data, and safety management.

[0055] In some embodiments, the push interface supports JSON-RPC or RESTful protocols and can be seamlessly integrated with scheduling systems, equipment control systems and other industrial information systems.

[0056] In other embodiments, the pushed content also includes the original text of regulatory clauses, risk levels, timestamps, and suggested action instructions for downstream systems to execute automatically. The workflow management module is responsible for task scheduling and tool invocation, including natural language parsing tools, vector retrieval tools, logical expression generation tools, and visualization tools. Each module is executed collaboratively by an AI agent and connected via a bus or network.

[0057] In one specific embodiment: when the gas concentration exceeds the limit and is determined to be high risk, the system will push the matched Article 138 of the Coal Mine Safety Regulations, the structured interpretation result "gas concentration > 1.0%", the compliance status "violation", and the rectification suggestion "immediately stop the operation and increase ventilation" to the coal mine safety management platform through the RESTful interface. The platform will automatically trigger an audible and visual alarm and send an SMS notification to the safety officer.

[0058] Specifically, the closed-loop linkage refers to the complete closed-loop process from production data collection, regulatory matching, compliance determination, risk warning to rectification suggestion output, and then to the execution of control commands (such as automatic power off and start of ventilation equipment) by the safety management platform.

[0059] It should be noted that this embodiment links the regulatory judgment results with the safety management platform, realizing the transformation of regulatory knowledge from passive retrieval to active application, which significantly improves the automation level and response speed of coal mine safety production.

[0060] Please refer to Figure 2 , Figure 2 A structural block diagram of a coal mine regulation intelligent matching and structured interpretation device provided in an embodiment of the present invention; the specific device may include: The regulations acquisition module 100 is used to acquire coal mine regulations text and perform hierarchical splitting and metadata extraction. Semantic modeling module 200 is used to semantically encode the split legal clauses using a pre-trained language model to generate legal vectors and build a legal knowledge base containing legal vectors. The vector matching module 300 is used to acquire coal mine production scene text and perform standardization processing, generate production scene vectors using the same encoding model as the pre-trained language model, calculate the semantic similarity between the production scene vectors and each regulation vector in the regulation knowledge base, and filter out a set of candidate regulation clauses whose similarity meets the preset conditions. The structured interpretation module 400 is used to call a natural language parsing tool to perform structured parsing on each candidate legal clause in the candidate legal clause set, extract information on the work object, constraints, safety parameter thresholds and handling measures, and generate structured safety rules containing logical judgment expressions. The compliance determination module 500 is used to substitute real-time collected coal mine production data into the logical judgment expression in the structured safety rules to determine compliance, and generate and output risk level assessment and rectification suggestions based on the determination results.

[0061] In some embodiments, the device further includes a push module for pushing the matched regulatory number, structured interpretation results, compliance status, and rectification suggestions to the coal mine safety management platform. The device also includes a workflow management module responsible for task scheduling and tool invocation, including natural language parsing tools, vector retrieval tools, logical expression generation tools, and visualization tools. Each module is executed collaboratively by an AI agent and connected via a bus or network.

[0062] In other embodiments, the formula for generating the regulatory clause vector in the semantic modeling module 200 is as follows: ,in This represents the text of the i-th legal provision. This indicates a pre-trained Chinese legal language model. This represents the generated vector of legal clauses, with a dimension of 1024. The semantic modeling module 200 also employs a cosine similarity formula. Calculate semantic similarity between clauses to optimize the knowledge base structure, where Terms and Conditions With Terms semantic similarity, Representing vectors The norm of .

[0063] In one specific embodiment: the regulation acquisition module 100 acquires the text of the "Coal Mine Safety Regulations" from the national regulation database, splits it according to clauses, and outputs it to the semantic modeling module 200; the semantic modeling module 200 calls the Legal BERT model to generate regulation vectors and stores them in the Milvus database; the vector matching module 300 obtains production scene text from underground sensors and generates scene vectors. Then, cosine similarity matching is performed with the regulatory vector. Output a set of candidate regulatory clauses The structured data interpretation module 400 calls the SpaCy tool to extract structured fields and maps them using rule functions. Generate a structured field vector and use Boolean logic formulas. Construct logical judgment expressions; the compliance judgment module 500 will generate data vectors in real time. Substitute expression Through compliance indicators Make a judgment and use the risk level assessment function. Output the risk level.

[0064] Specifically: the formula for calculating cosine similarity in the vector matching module 300 is as follows: ,in This represents the normalized and dimensionality-reduced production scene vector. This represents a vector of regulatory clauses. The calculation formula for the compliance indicators in the compliance determination module 500 is as follows: The expression for the risk level assessment function is: The specific judgment logic is as follows: when At that time, it was high-risk. At the time, it was considered a medium-risk period. The risk level is currently low.

[0065] It should be noted that the intelligent matching and structured interpretation device for coal mine regulations in this embodiment is used to implement the aforementioned intelligent matching and structured interpretation method for coal mine regulations. Therefore, the specific implementation methods of the intelligent matching and structured interpretation device for coal mine regulations can be found in the previous embodiment section of the intelligent matching and structured interpretation method for coal mine regulations. For example, the regulation acquisition module 100, semantic modeling module 200, vector matching module 300, structured interpretation module 400, and compliance judgment module 500 are respectively used to implement steps S101 to S104 in the above-mentioned intelligent matching and structured interpretation method for coal mine regulations. Therefore, the specific implementation methods can be referred to the descriptions of the corresponding embodiments, and will not be repeated here. In addition, the system of this device can run on a general server or industrial PC. The processor can be an Intel Xeon or ARM multi-core CPU, the memory configuration is not less than 16GB, and the storage capacity is not less than 1TB to support real-time regulation matching and production data processing.

[0066] A specific embodiment of the present invention also provides a coal mine regulations intelligent matching and structured interpretation device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-mentioned coal mine regulations intelligent matching and structured interpretation method.

[0067] A specific embodiment of the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described intelligent matching and structured interpretation method for coal mine regulations.

[0068] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for intelligent matching and structured interpretation of coal mine regulations, characterized in that, include: The text of coal mine regulations is obtained and hierarchically split and metadata is extracted. A pre-trained language model is used to semantically encode the split regulations to generate regulations vectors, and a regulations knowledge base containing regulations vectors is constructed. The text of a coal mine production scenario is obtained and standardized. A production scenario vector is generated using the same encoding model as the pre-trained language model. The semantic similarity between the production scenario vector and each legal vector in the legal knowledge base is calculated. A set of candidate legal clauses whose similarity meets the preset conditions is selected. The natural language parsing tool is invoked to perform structured parsing on each candidate regulation clause in the candidate regulation clause set, extracting information on the work object, constraints, safety parameter thresholds and handling measures, and generating structured safety rules containing logical judgment expressions; The real-time collected coal mine production data is substituted into the logical judgment expression in the structured safety rules to determine compliance. Based on the judgment result, a risk level assessment and rectification suggestions are generated and output.

2. The method according to claim 1, characterized in that, The process of acquiring coal mine regulations text, performing hierarchical decomposition and metadata extraction, using a pre-trained language model to semantically encode the decomposed regulations into regulations vectors, and constructing a regulations knowledge base containing these vectors includes: Coal mine safety-related regulatory texts are obtained from a multi-source regulatory platform, and standardized regulatory texts are obtained by unifying the encoding and formatting. The standardized regulatory text is split into chapters, clauses, and sentences, and the clause number, clause title, applicable objects, and regulatory level meta-information are extracted to obtain split clauses with meta-information. A pre-trained Chinese legal language model is invoked to semantically encode the clause text in the split clauses containing meta-information, generating a legal clause vector. The vectors of the aforementioned legal provisions are stored in a vector database to construct a legal knowledge base.

3. The method according to claim 1, characterized in that, The process of acquiring and standardizing coal mine production scenario text, generating production scenario vectors using the same encoding model as the pre-trained language model, calculating the semantic similarity between the production scenario vectors and the legal vectors in the legal knowledge base, and selecting a set of candidate legal clauses whose similarity meets preset conditions includes: Original production scene descriptions are obtained from multi-source data terminals at the coal mine production site, and noise reduction, word segmentation and standardization are performed to obtain a standardized production scene text set. The pre-trained language model is invoked to encode each production scenario description in the standardized production scenario text set, generating a production scenario vector. Normalization and dimensionality reduction are performed on the production scene vector to obtain the normalized and dimensionality-reduced scene vector; Calculate the cosine similarity between the normalized and dimensionality-reduced scene vector and each regulation vector in the regulation knowledge base, and select regulation clauses with similarity greater than or equal to the matching similarity threshold as a set of candidate regulation clauses.

4. The method according to claim 1, characterized in that, The step of calling a natural language parsing tool to perform structured parsing on each candidate regulation clause in the candidate regulation clause set, extracting information on the work object, constraints, safety parameter thresholds, and handling measures, and generating structured safety rules containing logical judgment expressions includes: The AI ​​agent invokes natural language processing tools to perform structured analysis, entity recognition, and semantic dependency parsing on candidate regulatory clauses, automatically extracting key information such as job type, constraints, safety thresholds, and handling measures. A rule mapping function is established based on the extracted key information to map candidate legal clauses into a structured field vector; For clauses containing complex combinations of conditions, a logical judgment expression is constructed using Boolean logic formulas. This logical judgment expression combines multiple condition judgment units through logical AND operations to determine whether the production status is compliant.

5. The method according to claim 1, characterized in that, The process involves substituting real-time collected coal mine production data into the logical judgment expression in the structured safety rules to determine compliance, and generating a risk level assessment and rectification suggestions based on the determination result, which are then output as follows: Acquire real-time coal mine production data and construct a real-time production data vector, then substitute the real-time production data vector into the logical judgment expression in the structured safety rules to calculate the truth value; Compliance indicators are generated based on the calculation results using indicator functions. These compliance indicators are used to indicate whether the production status complies with regulatory requirements or whether there are any violations. Based on the aforementioned compliance indicators and the degree of deviation between the real-time production data vector and the regulatory constraint thresholds, the risk level is calculated using a risk level assessment function, and corresponding rectification suggestions are generated based on the risk level.

6. The method according to claim 5, characterized in that, The calculation of the risk level based on the deviation between the compliance indicators and real-time production data vectors and regulatory constraint thresholds, using a risk level assessment function, includes: Extract key production parameters and corresponding regulatory constraint thresholds from real-time production data vectors; When multiple key production parameters simultaneously exceed the corresponding regulatory thresholds, the risk level is determined to be high. When any one of the key production parameters exceeds the corresponding regulatory threshold, it is classified as medium risk; When all key production parameters are less than or equal to the corresponding regulatory thresholds, the risk level is determined to be low.

7. The method according to claim 1, characterized in that, The method further includes: The matched regulation number, structured interpretation results, compliance status, and rectification suggestions are pushed to the coal mine safety management platform to achieve closed-loop linkage between regulatory provisions, production data, and safety management.

8. A device for intelligent matching and structured interpretation of coal mine regulations, characterized in that, include: The regulations acquisition module is used to acquire coal mine regulations texts and perform hierarchical splitting and metadata extraction. The semantic modeling module is used to semantically encode the split legal clauses using a pre-trained language model to generate legal vectors and build a legal knowledge base containing legal vectors. The vector matching module is used to acquire coal mine production scene text and perform standardization processing, generate production scene vectors using the same encoding model as the pre-trained language model, calculate the semantic similarity between the production scene vectors and each legal vector in the legal knowledge base, and filter out a set of candidate legal clauses whose similarity meets the preset conditions. The structured interpretation module is used to call natural language parsing tools to perform structured parsing on each candidate legal clause in the candidate legal clause set, extract information on the work object, constraints, safety parameter thresholds and disposal measures, and generate structured safety rules containing logical judgment expressions; The compliance determination module is used to substitute real-time collected coal mine production data into the logical judgment expression in the structured safety rules to determine compliance, and generate and output risk level assessment and rectification suggestions based on the determination results.

9. A device for intelligent matching and structured interpretation of coal mine regulations, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the intelligent matching and structured interpretation method for coal mine regulations as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the intelligent matching and structured interpretation method for coal mine regulations as described in any one of claims 1 to 7.

Images