Dangerous chemical substance electronic work ticket examination system and method based on large model analysis

Abstract

The invention discloses a hazardous chemical substance electronic work ticket examination system and method based on large model analysis, and the method comprises the following steps: 1, receiving and managing an electronic work ticket file, building a unique identifier for each work ticket, and tracking the examination state of each work ticket; 2, performing information extraction on the electronic work ticket file, and identifying and structuring key information fields including work content, work place, risk identification, safety measures and work time; and 3, inputting the structured key information field into a pre-trained large model compliance rule base for multi-dimensional compliance auditing, wherein the large model compliance rule base is constructed by learning massive labeled compliance and non-compliance job ticket samples and related law and regulation texts. According to the method, the auditing efficiency and accuracy can be improved, the rule adaptability is optimized, rectification closed-loop management is realized, the data value is mined, safety management is promoted to be converted into beforehand prevention, and the auditing capability of the system can be continuously optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of review systems, in particular to a dangerous chemical electronic work ticket review system and method based on large model analysis. BACKGROUND

[0002] The current dangerous chemical electronic work ticket review mainly relies on manual review, supplemented by basic information system to complete file archiving and simple information input. Enterprise personnel submit work tickets (formats include PDF, pictures, structured forms, etc.) of types such as ignition by offline filling or online form input. The reviewer needs to manually check relevant regulations and standards and enterprise internal systems to determine the compliance of key information such as ignition level, work site, safety measures, etc. in the work ticket. After review, the reviewer manually feeds back rectification opinions. The system only records the review results and rectification situation. Although some systems introduce OCR technology for simple character recognition, they cannot conduct in-depth analysis combined with industry rules, nor form a full-process closed-loop management.

[0003] The existing technology has obvious defects: first, the review efficiency is low, manual review takes a long time and has high labor cost, which is difficult to meet the demand of large number of work ticket review and the timeliness requirement of dangerous chemical work; second, the review accuracy is insufficient, which is highly dependent on the professional ability and industry experience of the reviewer, and is prone to inconsistent review standards and missed identification of deep compliance issues due to differences in personnel experience, and is difficult to avoid logical contradictions and details; third, the rule adaptability is poor, the fixed rule configuration mode lacks dynamic updating mechanism, which cannot match the changes of regulations and standards and enterprise management needs, and is difficult to adapt to the individualized review focus of dangerous chemical enterprises of different scales and different processes; fourth, the rectification tracking is not closed-loop, the rectification suggestions are general and lack operability, the rectification effect needs to be manually checked, and there is no rectification progress tracking and early warning function, the "review-rectification-recheck" process is broken; fifth, the data utilization value is low, the work ticket is stored in unstructured format, the key information is not extracted, and there is lack of data mining and deep analysis capability, which cannot provide support for safety management decision-making. The above-mentioned deficiencies of the existing technology make it difficult to guarantee the safety, efficiency and accuracy of the dangerous chemical electronic work ticket review. Therefore, a dangerous chemical electronic work ticket review system and method based on large model analysis are proposed. SUMMARY

[0004] The present application solves the above technical problems by the following technical solutions, which comprises the following steps: Step one: receiving and managing electronic work ticket files, establishing a unique identifier for each work ticket and tracking its review status; Step two: information extraction on the electronic work ticket files, identifying and structuring key information fields such as work content, work site, risk identification, safety measures and work time; Step three: input the structured key information field into the pre-trained large model compliance rule base for multi-dimensional compliance audit, and the large model compliance rule base is constructed by learning a large number of labeled compliant and non-compliant work order samples and related regulation texts; Step four: according to the audit result of step three, if the work order is not compliant, generate targeted rectification suggestions and feedback, then receive the rectified work order and return to step three for verification until the audit is passed; if the work order is compliant, it is approved and archived; Step five: based on historical audit and rectification data, statistical analysis of high-frequency violation patterns, and continuous optimization of the large model compliance rule base using the data.

[0005] Further, the information extraction of the electronic work order file in step two specifically includes: Using optical character recognition technology to convert the picture or PDF format work order into raw text information; Based on natural language processing technology, identifying and extracting predefined entity fields from the raw text information, including at least: work type, fire level, equipment number, risk point description and safety measure item; Map and fill the extracted unstructured entity field values into the corresponding structured fields according to the preset dangerous chemical work order data model, and generate standardized work order data objects.

[0006] The multi-dimensional compliance audit in step three is performed by the large model compliance rule base, which specifically includes the compliance score calculation and judgment process based on the output probability of the embedded large model: Set the rule set for audit as R={r1,r2,...,rn}, which contains n predefined audit rules, each rule r i is associated with a predefined confidence weight w i and a rule threshold T i ; Input the structured key information field into the large model compliance rule base, and each rule r i in the rule set R is independently judged by the embedded large model of the large model compliance rule base, and a probability value p i indicating the compliance degree of the rule is output; Based on the confidence weight w i , the probability values pi output by all rules in the rule set R are weighted and aggregated to calculate the comprehensive compliance score S of the work order; The calculation of the comprehensive compliance score S is a weighted average process of the output probability of all elements in the rule set R, and the specific formula is as follows: ; wherein pi is in the range of [0, 1]; The comprehensive compliance score S is compared with a preset compliance threshold value, and if S is lower than the compliance threshold value, an audit result is generated, and the audit result is that the work order is determined to be non-compliant.

[0007] Further, in the step three, if the work order is determined to be non-compliant, further execution of violation location and severity classification is performed: All compliance probabilities p are identified by traversing the rule set R i that are lower than their corresponding rule threshold values T i , and are marked as triggered violation items; According to a preset mapping relationship between rule types and severity levels, the severity level L i of each triggered violation item is determined, and the severity level includes critical violation, important violation, and general violation; Based on the severity levels of all triggered violation items, a total violation score V is calculated, which is used to quantify the degree of non-compliance of the current work order, and the calculation method is: ; wherein m is the number of types of violation severity levels, k j is the severity coefficient corresponding to the jth type of violation severity level, c j is the number of occurrences of the jth type of violation severity level according to the triggered violation items; The triggered violation items, their corresponding violation severity levels L i and the total violation score V are integrated to generate and output the audit detail data of the current work order, which is part of the audit result and is used to describe the non-compliant situation in detail.

[0008] Further, in the step four, specific rectification suggestions are generated, specifically including: According to the audit detail data, including the total violation score V, triggered violation items and their violation severity levels, the case with the highest similarity is retrieved from the historical rectification case library, and the similarity is calculated based on the matching degree of violation type, severity level and work scene; The violation information of the current work order is compared with the retrieved case, and the effective rectification measures in the case are extracted; The extracted rectification measures are adaptively modified and supplemented in combination with the requirements of the corresponding rules in the large model compliance rule library, to generate a rectification suggestion report containing specific modification fields, modification basis and operation guidelines; In the rectification suggestion report, the items determined as key violations are prioritized and highlighted according to the severity level of the triggered violations. The suggestions corresponding to these key violations are prioritized and highlighted. According to the audit details data, including the overall violation total score V, the triggered violation items and their violation severity levels, the most similar case is retrieved from the historical rectification case library based on the matching degree of violation type, severity level and work scene. Compare the violation information of the current work order with the retrieved case to extract effective rectification measures from the case. Combine the requirements of the corresponding rules in the large model compliance rule library to adaptively modify and supplement the extracted rectification measures, and generate a rectification suggestion report containing specific modification fields, modification basis and operation instructions. In the rectification suggestion report, the items determined as key violations are prioritized and highlighted according to the severity level of the triggered violations.

[0009] Further, in step four, the rectified work order is verified, including quantitative rectification effect evaluation: Input the key information fields of the rectified work order into the large model compliance rule library again, and perform audit by the large model embedded in the large model compliance rule library to obtain new audit results, including new comprehensive compliance score And new overall violation total score ; Calculate the improvement degree of the scores before and after rectification : ; Where S and V are the comprehensive compliance score and overall violation total score before rectification, and a is a preset violation score conversion coefficient for balancing the relationship between compliance score improvement and violation reduction. If the score improvement degree is greater than the preset improvement threshold, and the new overall violation total score is zero, it is determined that the rectification verification is passed. Otherwise, it is determined that the rectification is not complete, and supplementary rectification suggestions are generated based on the uneliminated violation items.

[0010] Further, in step five, the large model compliance rule library is continuously optimized, including dynamically updating the confidence weight of the embedded large model: Record each rule compliance probability p i output by the large model embedded in the large model compliance rule library in each audit icorresponding relationship, wherein the real compliance state y i Through artificial review or verification and confirmation of the rectified operation ticket; For each rule r i , based on the corresponding relationship, periodically calculate its recent prediction accuracy A i ; According to the prediction accuracy A i , adjust its confidence weight w i , the update formula is: ; Where β is a preset learning rate coefficient, A base is a preset baseline accuracy; The update process produces an optimized confidence weight set.

[0011] Further, the step five of statistical analysis of high-frequency violation patterns specifically includes: Periodically cluster analysis on the audit details data of historical non-compliant operation tickets, identify the combination of trigger violation items that often appear at the same time, and form high-risk violation patterns; Calculate the occurrence frequency and average overall violation total score of each of the high-risk violation patterns in the statistical period; High-risk violation patterns with occurrence frequency higher than a first preset threshold and average overall violation total score higher than a second preset threshold, and their corresponding typical operation scene information, are summarized and output to form a risk analysis report.

[0012] A dangerous chemical product electronic operation ticket examination system based on large model analysis, the system comprises: An operation ticket management module for receiving and managing electronic operation ticket files, establishing a unique identifier for each operation ticket and tracking its audit status; An information extraction and structuring module for extracting information from the electronic operation ticket files, identifying and structuring key information fields including operation content, operation location, risk identification, safety measures and operation time; An intelligent compliance audit module integrated with a pre-trained large model compliance rule library for receiving the key information fields and performing multi-dimensional compliance audit, the large model compliance rule library is constructed by learning a large number of labeled compliant and non-compliant operation ticket samples and related regulation texts, and embedded with a large model for performing rule reasoning and judgment; A rectification and verification closed-loop module for generating targeted rectification suggestions and feedback if the operation ticket is non-compliant according to the audit results of the intelligent compliance audit module, then receiving the rectified operation ticket and triggering the intelligent compliance audit module to re-verify until the audit is passed; if the operation ticket is compliant, trigger the approval archiving operation; The data analysis and optimization module is used to statistically analyze high-frequency violation patterns based on historical audit and rectification data, and to continuously optimize the compliance rules of the large model using the data.

[0013] Compared with existing technologies, this invention has the following advantages: The hazardous chemicals electronic work permit review system and method based on large-scale model analysis automatically receives and manages work permits, establishing unique identifiers and status tracking mechanisms. Combined with structured extraction of key information fields and large-scale model batch review, it significantly reduces manual intervention, solving the problems of time-consuming manual review and backlog in batch processing, thus meeting the timeliness requirements of hazardous chemicals operations. Relying on a large-scale model compliance rule base, it learns from massive amounts of compliant and non-compliant work permit samples and relevant regulatory texts to conduct multi-dimensional compliance reviews of work permits, including basic information verification, logical relationship verification, and regulatory clause matching. This accurately identifies non-compliant fields and problem types, avoiding the shortcomings of manual review caused by experience differences, inconsistencies in standards, and insufficient identification of deep-seated problems. The model compliance rule base has dynamic optimization capabilities, continuously iterating based on newly added regulatory texts, work permit samples, and historical review data. It can adapt to the personalized management needs of hazardous chemical enterprises of different sizes and with different processes, and keep up with changes in regulations and standards in a timely manner, solving the problems of rigid rules, lagging updates, and poor adaptability of traditional systems. For non-compliant work orders, it generates targeted and actionable rectification suggestions by combining audit details data and historical rectification cases, and evaluates the rectification effect through automated verification, forming a closed loop of the entire process from review to rectification to review. At the same time, it tracks the rectification progress and can provide timely feedback on incomplete rectification and overdue rectification, solving the problems of general rectification suggestions, manual verification, and process breaks in traditional systems. By extracting and transforming key information from work orders into standardized structured data, and combining it with data analysis modules to statistically analyze high-frequency violation patterns and identify high-risk violation combinations, it can explore the correlation between work order data and safety hazards, providing support for safety management decisions, promoting the transformation of safety management from post-event rectification to pre-event prevention, and solving the problems of low data storage and archiving and low utilization value of traditional technical data. Based on historical audit data, rectification cases, and manual review results, the confidence weight and audit logic of the large model compliance rule base are dynamically adjusted to continuously improve the system's audit accuracy and adaptability, ensuring that the system can continuously optimize performance as the use scenarios expand, making the system more worthy of promotion and use. Attached Figure Description

[0014] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0015] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0016] like Figure 1 As shown, this embodiment provides a technical solution: a method for reviewing electronic work permits for hazardous chemicals based on large model analysis, including the following steps: Step 1: Receive and manage electronic work order files, establish a unique identifier for each work order, and track its review status; Step 2: Extract information from the electronic work order file, identify and structure key information fields including work content, work location, risk identification, safety measures and work time; Step 3: Input the structured key information fields into the pre-trained large model compliance rule library for multi-dimensional compliance review. The large model compliance rule library is constructed by learning from a large number of labeled compliant and non-compliant work ticket samples and relevant regulatory texts. Step 4: Based on the review results of Step 3, if the work ticket is non-compliant, generate targeted rectification suggestions and provide feedback. Then, receive the rectified work ticket and return to Step 3 for verification until the review is passed; if the work ticket is compliant, approve and archive it. Step 5: Based on historical audit and rectification data, statistically analyze high-frequency violation patterns, and use the data to continuously optimize the large model compliance rule library.

[0017] The information extraction process for the electronic work order file in step two specifically includes: Optical character recognition technology is used to convert work tickets in image or PDF format into raw text information; Based on natural language processing technology, predefined entity fields are identified and extracted from the original text information. The entity fields include at least: work type, hot work level, equipment number, risk point description and safety measure items. The extracted unstructured entity field values ​​are mapped and filled into the corresponding structured fields according to the preset hazardous chemical work order data model, generating standardized work order data objects. The pre-defined hazardous chemicals operation ticket data model is a core sub-module (data standardization sub-module) of the information extraction and structuring module. Essentially, it is a structured data standard system customized for hazardous chemicals operation scenarios, comprising three core components: Field Classification Definition: Clearly define 3 categories and 18 mandatory structured fields, including: basic work information (work type, work location, work time, etc.), risk control (risk point description, safety measure items, hot work level, etc.), and compliance verification (equipment number, approval status, sampling analysis interval, etc.). Data format specifications: Define the data type (string, date and time, number, list, etc.) and format requirements for each field (e.g., the work time field must follow the format "YYYY-MM-DDHH:MM", and the hot work level field is only allowed to take the values ​​"Level 1 / Level 2 / Special Level"). Field association rules: Clearly define the logical relationships between fields (e.g., when "the work location is a major hazard source area," the "hot work rating" and "combustible gas monitoring measures" fields must be associated), providing a data association foundation for subsequent compliance audits. This submodule has a built-in extensible interface that supports updating field definitions based on the process characteristics and industry standards of different hazardous chemical companies; By standardizing information extraction and structuring electronic work ticket files, unstructured work ticket data is transformed into standardized work ticket data objects. This solves the problems of disorganized work ticket information storage, reliance on manual extraction of key information which is prone to errors and omissions, and inconsistent data formats leading to low efficiency in subsequent review and difficulties in data utilization in existing technologies. It provides accurate and unified data source support for multi-dimensional compliance review of the subsequent large-scale model compliance rule base, while improving the traceability and analyzability of work ticket data, laying the foundation for subsequent data mining and safety management decisions.

[0018] Suppose a hazardous chemical company uploads an electronic work order in PDF format, containing the following information: "Work Type: Hot Work, Hot Work Rating: Level 1, Work Location: Next to Equipment A01 in Workshop A, Risk Description: There is a risk of flammable gas leakage in the work area, Safety Measures: One dry powder fire extinguisher on site, operators must be certified, Work Time: 8:00 AM to 6:00 PM on October 20, 2025, Equipment Number: A01". First, this PDF work order is converted into the aforementioned raw text information. Then, predefined entity fields are extracted from the raw text to obtain: "Work Type = Hot Work, Hot Work Rating = Level 1, Equipment Number = A01, Risk Description = There is a risk of flammable gas leakage in the work area, Safety Measures = One dry powder fire extinguisher on site, operators must be certified, Work Time = 8:00 AM to 6:00 PM on October 20, 2025". Finally, according to a pre-defined hazardous chemical work order data model, the unstructured entity field values ​​are mapped and filled into the corresponding structured data. In the fields, a standardized work ticket data object is generated. The "Work Time" field is split into "Work Start Time = October 20, 2025, 08:00" and "Work End Time = October 20, 2025, 18:00". The remaining entity fields are respectively filled into structured fields such as "Work Type", "Hot Water Level", "Equipment Number", "Risk Point Description", and "Safety Measures Item", forming standardized data with a unified format and complete information. This ensures that the large model compliance rule base can quickly and accurately obtain key information for compliance comparison in the subsequent step three.

[0019] The multi-dimensional compliance review in step three is executed by the large model compliance rule base, specifically including the compliance score calculation and judgment process based on the output probability of its embedded large model: Let the set of rules used for auditing be R = {r1, r2, ..., rn}, which contains n predefined auditing rules, each rule r i Associated with a predefined confidence weight w i and a rule threshold T i ; The structured key information fields are input into the large model compliance rule base, and the embedded large model applies each rule r in the rule set R. i Perform an independent judgment and output a probability value p representing the degree of compliance of the rule. i ; The large-scale compliance rule base is a core sub-module (rule storage and scheduling sub-module) of the intelligent compliance audit module, providing underlying rule support for compliance audits. Its specific components include: Rule set storage unit: Stores n preset audit rules (R={r1,r2,...,rn}), covering three categories: basic information verification rules, logical relationship verification rules, and legal clause matching rules. Each rule is associated with a unique identifier, confidence weight, rule threshold, and applicable scenario tag. Regulatory Text Knowledge Base: Integrates industry regulations such as "GB30871 Safety Specifications for Special Operations of Hazardous Chemical Enterprises" and internal safety management systems of enterprises to form structured regulatory entries, providing a basis for rule matching; Violation Level Mapping Table: The correspondence between preset rule types and violation severity levels (critical violation / important violation / general violation) and severity coefficients; Rule scheduling interface: Receives structured key information fields, calls the embedded large model to execute rule judgments, and returns compliance probability values ​​and triggered violation items. This submodule supports manual rule addition, batch import, and automatic updates based on data analysis; Based on the confidence weight w i The probability values ​​pi output by all rules in the rule set R are weighted and aggregated to calculate the comprehensive compliance score S of the work ticket; The comprehensive compliance score S is calculated by weighting the output probabilities of all elements in the rule set R, and the specific formula is as follows: ; Wherein, the value range of pi is [0,1]; The overall compliance score S is compared with the preset compliance threshold. If S is lower than the compliance threshold, an audit result is generated, and the audit result is that the work order is deemed non-compliant. The embedded large model is the core sub-module for inference and computation of the intelligent compliance audit module. It is built on the Transformer architecture and has been fine-tuned and optimized using data from the field of hazardous chemical operations compliance. It has three core capabilities: Rule matching reasoning: Receives structured key information and a set of rules from a large model compliance rule library, independently judges the compliance level of each rule, and outputs the probability value in the range [0,1]. Probability-weighted aggregation: Based on the rule confidence weights, the compliance probabilities of multiple rules are weighted and calculated to generate a comprehensive compliance score; Violation Logic Analysis: Identifies the relationships between triggering violations to assist in determining the severity level of violations. This submodule's training data includes over 100,000 labeled compliant / non-compliant work order samples and over 5,000 industry regulatory texts. The model has 1.3 bytes of parameters and an inference latency of ≤500ms, ensuring both efficiency and accuracy in the review process. By using a large-scale model compliance rule base to perform multi-dimensional compliance audits on structured key information, and combining predefined confidence weights with rule compliance probability values ​​to calculate a comprehensive compliance score, compliance can be quantitatively determined. This solves the problems of inconsistent manual audit standards, insufficient identification of deep-seated compliance issues, and inability to accurately measure the degree of compliance, improving the objectivity, accuracy, and standardization of audits. At the same time, it provides a quantitative basis for subsequent rectification priority determination and avoids compliance misjudgments or omissions caused by subjective human judgment.

[0020] Define a set of review rules R = {r1, r2, r3}, where: r1 (Basic Rule - Required Field Validation): The approver's signature cannot be empty, confidence weight w1=0.3, rule threshold T1=0.8; r2 (Logical Rule - Hot Work Rating Matching): When the work location is a major hazard source area (flammable and explosive area), the hot work rating must be Special Grade, with confidence weight w2=0.4 and rule threshold T2=0.85; r3 (regulations and rules - sampling and analysis interval): sampling and analysis interval ≤ 30 minutes (according to GB30871 Safety Specifications for Special Operations of Hazardous Chemical Enterprises), confidence weight w3 = 0.3, rule threshold T3 = 0.9; The preset compliance threshold is 0.85.

[0021] Suppose that a hot work permit uploaded by a hazardous chemical company, after processing, has the following structured key information: "Approver Signature: Zhang X, Hot Work Level: Level 1, Work Location: Next to Equipment A01 in Workshop A (Major Hazard Source Area), Sampling and Analysis Interval: 40 minutes". This structured information is input into the compliance rule base of a large-scale model. The embedded model independently judges each rule and outputs a probability value. r1: The approver's signature is complete, and the compliance probability p1=0.95 (meets the requirement pi∈[0,1]). r2: The hot work level is Level 1, which does not meet the requirement of special-level hot work in major hazard source areas. The compliance probability p2=0.7; r3: The sampling and analysis interval of 40 minutes exceeds the regulatory requirement, and the compliance probability p3=0.6.

[0022] Calculated according to the comprehensive compliance scoring formula: ; Comparing S=0.745 with the preset compliance threshold of 0.85, since 0.745 < 0.85, the work ticket is determined to be non-compliant.

[0023] In step three, if the work order is determined to be non-compliant, further steps are taken to locate the violation and classify its severity: Traverse the rule set R to identify all compliance probabilities p i Below its corresponding rule threshold T i The rule ri is then marked as a violation. Based on the pre-defined mapping relationship between rule types and severity levels, the severity level L of each triggered violation is determined. i The severity levels of violations include critical violations, major violations, and minor violations; Based on the severity level of all triggering violations, an overall violation score V is calculated to quantify the degree of non-compliance of the current work order. The calculation method is as follows: ; Where m is the number of types of violations with varying degrees of severity, and k j c is the severity coefficient corresponding to the severity level of the j-th type of violation. j This refers to the number of occurrences belonging to the severity level of the j-th type of violation, calculated based on the aforementioned triggering violation items. The system integrates the triggered violation item, its corresponding violation severity level Li, and the overall violation score V to generate and output the audit details data of the current work order. This audit details data serves as a component of the audit results and is used to describe the non-compliance in detail. After determining that the work ticket is non-compliant, the system accurately locates the triggering violation, classifies the severity of the violation according to the rule type, and calculates the overall violation score. This enables precise and quantitative management of the violation issues, which solves the shortcomings of existing technologies such as vague violation descriptions, inability to measure the severity, and chaotic rectification priorities. It provides a clear basis for subsequent targeted rectification and priority ranking, avoids key violations from being overlooked, and improves the targeting and efficiency of rectification.

[0024] Based on a unified set of review rules and work order information, the following settings will continue: The audit rule set R = {r1, r2, r3}, and the thresholds and weights of each rule are the same as before: r1 (basic rule - mandatory item verification): T1 = 0.8, corresponding to general violations; r2 (logical rule - matching of firework levels, based on "the firework level needs to match the risk level of the operation location"): T2 = 0.85, corresponding to critical violations; r3 (regulatory rule - sampling analysis interval, based on "Safety Specification for Special Operations in Hazardous Chemical Enterprises GB30871"): T3 = 0.9, corresponding to major violations; Preset mapping relationship of violation severity levels: Critical violation (k1 = 3), Major violation (k2 = 2), General violation (k3 = 1), and the number of types of violation severity levels m = 3; After multi-dimensional compliance audits of this work permit, the compliance probabilities of each rule are: p1 = 0.95, p2 = 0.7, p3 = 0.6.

[0025] The method performs the following operations: Violation location: Traverse the rule set R to identify the triggered violation items where pi < Ti: r2: p2 = 0.7 < T2 = 0.85, marked as a triggered violation item; r3: p3 = 0.6 < T3 = 0.9, marked as a triggered violation item; r1: p1 = 0.95 ≥ T1 = 0.8, not constituting a violation.

[0026] Severity grading: Determine the severity level of the triggered violation items according to the mapping relationship: r2 (logical rule violation) → Critical violation (Li = Critical violation); r3 (regulatory rule violation) → Major violation (Li = Major violation).

[0027] Calculation of the overall violation total score: Count the occurrence times of various violations: Critical violation occurs 1 time (c1 = 1), Major violation occurs 1 time (c2 = 1), General violation occurs 0 times (c3 = 0), and substitute into the formula: ; Storage of audit details data: Store the triggered violation items (r2, r3), the corresponding severity levels (Critical violation, Major violation), and the overall violation total score V = 5.

[0028] The generation of targeted rectification suggestions in step four specifically includes: Based on the audit details data, including the overall violation score V, the triggering violation items and their severity levels, the case with the highest similarity is retrieved from the historical rectification case database. The similarity is calculated based on the matching degree of violation type, severity level and work scenario. The violation information of the current work order is compared with the retrieved cases, and effective rectification measures are extracted from the cases; Based on the requirements of the corresponding rules in the large model compliance rule library, the extracted rectification measures are adaptively modified and supplemented to generate a rectification suggestion report containing specific modified fields, modification basis and operation guidelines; In the rectification suggestion report, the items are sorted and marked according to the severity level of the violations. Items identified as critical violations will be prioritized and highlighted, and targeted rectification suggestions will be generated in step four. Specifically, these include: Based on the audit details data, including the overall violation score V, the triggering violation items and their severity levels, the case with the highest similarity is retrieved from the historical rectification case database. The similarity is calculated based on the matching degree of violation type, severity level and work scenario. The violation information of the current work order is compared with the retrieved cases, and effective rectification measures are extracted from the cases; Based on the requirements of the corresponding rules in the large model compliance rule library, the extracted rectification measures are adaptively modified and supplemented to generate a rectification suggestion report containing specific modified fields, modification basis and operation guidelines; In the rectification suggestion report, the suggestions are sorted and marked according to the severity level of the violations that triggered them. The suggestions corresponding to the items that are identified as key violations will be prioritized and highlighted. Based on the work order review details, by searching similar historical rectification cases, extracting effective measures, and combining them with the requirements of the large model compliance rule library, targeted rectification suggestions are generated, including specific modification fields, modification basis, and operation guidelines. Key violations are also marked and sorted according to the severity of the violation. This solves the problems of existing technical rectification suggestions being vague, lacking operability, and having chaotic rectification priorities. It allows users to accurately focus on core violations and rectify them efficiently, while providing clear standards for subsequent rectification verification, improving the quality of rectification and the efficiency of closed-loop management.

[0029] Continue using the standardized work order review details data: Triggering violations: r2 (critical violation, hot work level is level 1, the work location is next to equipment A01 in workshop A which is a major hazard source area, violating the logical rule that "the hot work level must match the risk level of the work location") and r3 (significant violation, sampling and analysis interval is 40 minutes, violating the regulation rule of "sampling and analysis interval ≤ 30 minutes" in "GB30871 Safety Specifications for Special Operations of Hazardous Chemical Enterprises"). Overall violation score V=5; The historical rectification case database stores multiple sets of rectification cases of violations of hot work operations involving hazardous chemicals. The case matching dimensions include violation type (logical violation / regulatory violation), severity level (critical / important / general), and operation scenario (hot work operation + major hazard source area).

[0030] Case Search: Based on the matching criteria of the current work order, namely "Violation Type = Logical Violation + Regulatory Violation, Severity Level = Critical Violation + Important Violation, Work Scenario = Hot Work + Major Hazard Source Area", the case with the highest similarity was retrieved from the historical rectification case database. The case information is as follows: "Work type: Hot work, Triggering violation: Hot work level 1 (critical violation, work location is next to equipment B02 in workshop B, major hazard source area), sampling and analysis interval 35 minutes (important violation), overall violation score V=4.5, rectification measures: 1. Adjust the hot work level to special level; 2. Shorten the sampling and analysis interval to 25 minutes to ensure compliance with the requirements of "GB30871 Safety Specifications for Special Operations of Hazardous Chemical Enterprises"; 3. Add real-time monitoring equipment for combustible gas concentration at the work site."

[0031] Measures Extraction and Optimization: Extract effective measures from the case, such as "adjusting the hot work level, shortening the sampling interval, and adding monitoring equipment", and make adaptive modifications and additions based on the corresponding rule requirements of the large model compliance rule library (r2 requires special hot work level, r3 sampling interval ≤ 30 minutes). Clarify the modified fields as "hot work level" and "sampling analysis interval", and supplement the details of the operation guidelines.

[0032] Generate a rectification recommendation report: Sorted and labeled according to the severity of violations, with key violations receiving priority. The report content is as follows: Priority rectification (critical violation): Modify the field "Hot work level", the basis for modification is "Hot work level must match the risk level of the work location (special level hot work is required in areas with major hazard sources)", and the operation guide is "Reassess the risk of the work environment next to equipment A01 in Workshop A, adjust the hot work level from level one to special level, and have it reviewed and confirmed by the enterprise's safety management department". Secondary rectification (major violation): Modify the field "sampling analysis interval" according to the requirement of "sampling analysis interval ≤ 30 minutes" in "GB30871 Safety Specifications for Special Operations of Hazardous Chemical Enterprises". The operation guide is "re-sampling and analysis, shorten the sampling interval to within 30 minutes, and retain the sampling record and analysis report for future reference". Additional suggestion: In light of the operational risks in areas with major hazard sources, install real-time monitoring equipment for combustible gas concentration at the work site and assign dedicated personnel to record the monitoring data every 15 minutes.

[0033] Step four involves verifying the rectified work orders, including a quantitative evaluation of the rectification effectiveness: The key information fields of the rectified work order are re-entered into the large-scale model compliance rule library, which then reviews the work order using the large-scale model embedded within the library, resulting in a new review result, including a new comprehensive compliance score. and the new overall violation score ; Calculate the improvement rate of the score before and after rectification. : ; Where S and V are the comprehensive compliance score and the overall violation score before rectification, and α is the preset violation score conversion coefficient, which is used to balance the relationship between the improvement of the compliance score and the reduction of violations. If the score improves The score exceeds the preset improvement threshold, and the new overall violation score is... If the value is zero, the rectification verification is deemed passed; Otherwise, if the rectification is deemed incomplete, supplementary rectification suggestions will be generated based on the remaining violations. By re-executing the compliance audit of the rectified work permits using a large-scale model, a new comprehensive compliance score and overall violation score are obtained. The rectification effect is then quantitatively evaluated using a score improvement formula. This solves the problems of existing technologies relying on manual verification for rectification, inaccurate measurement of effects, and difficulty in identifying incomplete rectification. At the same time, through clear verification standards (improvement threshold, requirement to clear the total violation score) and a supplementary rectification suggestion mechanism, it ensures that the rectification meets compliance requirements, improves the closed-loop management from review to rectification to re-verification, avoids leaving behind violations, and enhances the quality of closed-loop control in the review of hazardous chemical work permits.

[0034] Key data before rectification: Comprehensive compliance score S=0.745, overall violation score V=5; Preset parameter description: The violation score conversion factor α = 0.2 (set based on the actual need in the safety management of hazardous chemical operations that "improving compliance scores and eliminating violations are equally important", and is used to balance the weight of the two). Improvement threshold = 0.15 (based on historical rectification data statistics, set as the minimum standard for "significant improvement in compliance score and zero violations"); The set of audit rules R = {r1, r2, r3} (rule content, confidence weight w) i Rule threshold T i (Consistent with previous statements).

[0035] Work order information after rectification (the specific status after rectification is made public, meeting the requirements of full disclosure).

[0036] The user completed the adjustments based on the rectification suggestions. After the rectification, the uploaded hot work permit was structured and the key information is: "Approver's signature: Zhang X San, Hot work level: Special level (adjusted to the level corresponding to the major hazard source area as required), Work location: next to equipment A01 in workshop A (major hazard source area), Sampling and analysis interval: 25 minutes (shortened to the ≤30 minutes required by regulations)".

[0037] Post-rectification compliance audit and data calculation: Step 1: Re-enter the large model compliance rule library for review. The embedded large model independently judges each rule and outputs the compliance probability (based on reasonable inferences from the information after rectification). r1 (Approver's signature required): Compliance status unchanged. ; R2 (Hot Work Rating Matching): After rectification, the hot work rating is now classified as Special Grade, meeting the requirements for areas with major hazard sources. ; r3 (Sampling and analysis interval compliance): After rectification, the interval is 25 minutes to 30 minutes, which complies with "GB30871 Safety Specifications for Special Operations of Hazardous Chemical Enterprises". .

[0038] Step 2: Calculate the comprehensive compliance score after rectification. : .

[0039] Step 3: Calculate the overall violation score after rectification. : Traverse the rule set R, Since no violations were triggered, the number of violations of each type is c1=c2=c3=0. Substituting this into the formula: ; Improvement in rating Calculation and verification judgment; Substitute into the improvement formula: ; Verification judgment: Improvement in rating =1.196 > preset improvement threshold 0.15; New overall violation score ; If both conditions are met, the rectification verification is deemed successful, and the work order is filed in compliance with regulations.

[0040] If the user only performs partial rectification, and the post-rectification work order information is "Hot work level: Special, Sampling analysis interval: 35 minutes", then: Probability of compliance after rectification: (r3 is still in violation); calculate , (Only R3 is a major violation) ); Improvement in rating: ; although ,but The rectification was deemed incomplete, and supplementary rectification suggestions were generated: "Please shorten the sampling and analysis interval to within 30 minutes and retain the sampling records for future reference."

[0041] Step five involves continuously optimizing the large model compliance rule base, including dynamically updating the confidence weights of its embedded large models: Record the compliance probabilities p of each rule output by the large model embedded in the large model compliance rule library during each review. i With the final confirmed true compliance status y i The correspondence, where the actual compliance status y i Verification and confirmation can be conducted through manual review or by checking the rectified work orders. For each rule ri, its recent prediction accuracy A is periodically calculated based on the aforementioned correspondence. i ; Based on the prediction accuracy A i Adjust its confidence weight w i The updated formula is: ; Where β is the preset learning rate coefficient, and A base This is the preset baseline accuracy. This update process produces an optimized set of confidence weights; By recording the correspondence between rule compliance probability and actual compliance status, and periodically calculating the rule prediction accuracy and dynamically updating the confidence weight based on this accuracy, this technology solves the problems of fixed rule weights, inability to optimize based on actual audit results, underutilization of high-accuracy rules, or interference with audit results by low-accuracy rules in existing technologies. This ensures that the weight configuration of the large-scale compliance rule library continuously aligns with actual application scenarios, ensuring that high-value rules occupy a more critical position in compliance scoring, thereby continuously improving the audit accuracy and adaptability of the rule library and avoiding long-term audit bias caused by fixed weights.

[0042] The set of audit rules R = {r1, r2, r3} (rule content, initial confidence weight) (The rule threshold Ti remains the same as before): r1 (Basic rule - approver signature cannot be empty): ; r2 (Logic Rule - The level of hot work must match the risk level of the operation location): ; r3 (Regulatory Rule - Sampling analysis interval ≤ 30 minutes, in accordance with "GB30871 Safety Specification for Special Operations in Hazardous Chemical Enterprises"): ; Preset parameter description: Learning rate coefficient β = 0.1 (set based on the mild optimization requirements of hazardous chemical operation review rules to avoid excessive weight fluctuations affecting review stability); Benchmark accuracy A base = 0.85 (set based on the average effective accuracy rate of hazardous chemical operation ticket review rules in the industry, as the benchmark threshold for weight adjustment); Definition of the true compliance status yi: y i = 1 indicates that the true compliance status of rule ri is "compliant", y i = 0 indicates "non - compliant" (confirmed through manual review or verification after rectification); Statistical period: Select 100 recent complete review cases (including manual review / verification results after rectification) for calculating the prediction accuracy A of each rule i .

[0043] Historical review data statistics: Extract the compliance probability p of each rule r i from the 100 review data recorded by the system i and the corresponding relationship with the true compliance status y i . Count the number of times "the prediction result is consistent with the true state" and calculate the prediction accuracy A i (A i = Number of accurate predictions / Total number of statistical cases): r1 (Approver's signature is required): In 100 cases, the determination of p i is consistent with y i 88 times (e.g., when p i ≥ T1 = 0.8, y i = 1, when pi < T1, y i = 0), prediction accuracy A1 = 88 / 100 = 0.88; r2 (Match of hot work level): In 100 cases, the determination of p i is consistent with y i 92 times (e.g., when pi ≥ T2 = 0.85, yi = 1), prediction accuracy A2 = 92 / 100 = 0.92; r3 (Compliance of sampling analysis interval): In 100 cases, the determination of p i is consistent with y i 82 times (e.g., when pi ≥ T3 = 0.9, y i=1), prediction accuracy A3=82 / 100=0.82.

[0044] Confidence weights are dynamically updated and calculated: Based on the weight update formula: ; Calculate the new weights for each rule separately: For r1: Substitute the values: ; Step-by-step calculation: ; ; final: (Retain four decimal places to ensure weight accuracy).

[0045] For r2: Substitute the values: ; Step-by-step calculation: ; ; final: .

[0046] For r3: Substitute the values: ; Step-by-step calculation: ; ; final: .

[0047] The updated weight set is The following optimization effects are achieved: The prediction accuracy is higher than the baseline r1 (A1=0.88>A). base r2 (A2=0.92>A) base The weight of each criterion has been slightly increased, with the most significant increase in the weight of the high-accuracy r2 (hot work level matching, key safety logic rule), which will have a greater influence in subsequent compliance scoring. The prediction accuracy was lower than the baseline r3 (A3=0.82). base The weight is slightly reduced to minimize its interference with the overall review results and to leave room for subsequent rule optimization (such as adjusting the judgment logic of r3).

[0048] Step five, which involves statistically analyzing high-frequency violation patterns, specifically includes: Regularly perform cluster analysis on the audit details of historical non-compliant work tickets to identify frequently occurring combinations of triggering violations and form high-risk violation patterns; ​Calculate the frequency of occurrence and the average overall violation score for each of the aforementioned high-risk violation patterns within the statistical period; The high-risk violation patterns that occur more frequently than the first preset threshold and whose average overall violation score is higher than the second preset threshold, along with their corresponding typical work scenario information, are summarized and output to form a risk analysis report. By clustering analysis of historical non-compliant work order review details, this method accurately identifies frequently occurring combinations of triggering violations and forms high-risk violation patterns. Combined with a two-dimensional quantitative assessment of occurrence frequency and average overall violation score, it solves the problems of existing technologies being unable to uncover correlation patterns among violations, predicting potential high-risk work scenarios, and passive safety management. This provides precise and targeted basis for the safety management of hazardous chemical operations, helps to optimize management strategies in advance, strengthen the prevention and control of key scenarios, and promotes a deep transformation of safety management from "post-event rectification" to "pre-event prevention."

[0049] The set of review rules R = {r1, r2, r3} (the rule content, severity level, and severity coefficient are the same as above): r1 (Basic rule - approver signature cannot be empty) → General violation (k3=1); r2 (Logical rule - Hot work level must match the risk level of the work site) → Critical violation (k1=3); r3 (Regulations and rules - sampling and analysis interval ≤ 30 minutes, in accordance with GB30871 Safety Specifications for Special Operations of Hazardous Chemical Enterprises) → Major violation (k2=2); Preset parameter description: Statistical period: the past 6 months (covering both peak and off-peak seasons for hazardous chemical operations to ensure data representativeness); Total number of non-compliant work permits during the statistical period: N=200 (derived based on the average daily workload of a medium-sized hazardous chemical enterprise); First preset threshold (occurrence frequency threshold): Fth=20% (refer to the industry's high-risk mode judgment standard, that is, the occurrence frequency exceeds 1 / 5 and is considered high frequency). The second preset threshold (average overall violation score threshold): Vavg,th=4 (combined with the previous text on the range of violation scores V≥0, a score higher than 4.0 indicates a more serious violation). Clustering analysis logic: A clustering algorithm based on the similarity of violation item combinations is adopted to group work tickets that "simultaneously trigger the same set of violations" into the same pattern, and three main high-risk violation patterns are identified.

[0050] Historical data statistics and identification of high-risk violation patterns: By extracting the audit details data (triggering violations, overall violation score V) of 200 non-compliant work tickets within the statistical period, cluster analysis identified the following three high-frequency violation combination patterns: Mode 1: M1={r2,r3} (simultaneously triggering "Hot work level mismatch" + "Sampling analysis interval exceeds limit"), occurring 52 times. The overall violation score details for each violation include 5.0, 4.5, 5.0, ..., 4.0 (a total of 52 values), and the sum of all scores is V. sum1 =234; Mode 2: M2={r1,r3} (simultaneously triggering "Missing Approver Signature" + "Exceeding Sampling Analysis Interval"), occurring 30 times. The overall violation score details for each violation include 3.0, 2.5, 3.0, ..., 2.0 (a total of 30 values), and the sum of all scores is V. sum2}=75; Mode 3: M3={r1,r2,r3} (simultaneously triggers "missing approver signature" + "hot work level mismatch" + "excessive sampling analysis interval"), occurring 18 times. The overall violation score details for each violation include 6.0, 5.5, 6.0, ..., 5.0 (a total of 18 values), and the total score Vsum3=99.

[0051] Key performance indicator calculation: Calculate the occurrence frequency Fm of each pattern and the average overall violation score V{avg,m}: Formula for calculating the frequency of occurrence: (Cm represents the number of times the pattern occurred, and N represents the total number of non-compliant work tickets). Formula for calculating the average overall violation score: (Vsum,m is the sum of V values ​​for all violations of the pattern).

[0052] Calculate the indicators for each model separately: For M1: Occurrence frequency Average overall violation score: ; For M2: Occurrence frequency Average overall violation score: ; For M3: Occurrence frequency Average overall violation score: .

[0053] Output of high-risk violation patterns: Compare each model indicator with a preset threshold to select the " and The pattern: M1: Both thresholds are satisfied; M2: The occurrence frequency threshold is not met. M3: F3 = 9% < 20%, which does not meet the occurrence frequency threshold.

[0054] The final output includes information on high-risk violation patterns and typical operational scenarios that meet the requirements. High-risk violation patterns: ; Frequency of occurrence: 26% (26% of non-compliant work tickets within the statistical period simultaneously triggered this combination); Average overall violation score: 4.5 (relatively serious violation); Typical work scenarios: Hot work in areas with major hazard sources (such as next to equipment A01 in Workshop A or equipment area B02 in Workshop B); Control recommendations: For hot work operations in areas with major hazard sources, a new dual pre-approval process of "pre-approval of hot work level + special verification of sampling interval" should be added. Before the operation, the safety department and the technical department should jointly review the hot work level determination results, and the sampling analysis records should be uploaded to the system in real time to verify the interval time.

[0055] A hazardous chemicals electronic work permit review system based on large model analysis, the system comprising: The work ticket management module is used to receive and manage electronic work ticket files, establish a unique identifier for each work ticket, and track its review status. The information extraction and structuring module is used to extract information from the electronic work ticket file, identify and structure key information fields including work content, work location, risk identification, safety measures and work time; The intelligent compliance review module integrates a pre-trained large-scale model compliance rule library, which is used to receive the key information fields and perform multi-dimensional compliance review. The large-scale model compliance rule library is constructed by learning from a large number of labeled compliant and non-compliant work ticket samples and relevant regulatory texts, and has a large model embedded in it for performing rule reasoning and judgment. The rectification and verification closed-loop module is used to generate targeted rectification suggestions and provide feedback if the work ticket does not comply with the rules based on the audit results of the intelligent compliance audit module. Then, it receives the rectified work ticket and triggers the intelligent compliance audit module to re-verify until the audit is passed; if the work ticket complies with the rules, it triggers the approval and archiving operation. The data analysis and optimization module is used to statistically analyze high-frequency violation patterns based on historical audit and rectification data, and to continuously optimize the compliance rules of the large model using the data.

[0056] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0057] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0058] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for reviewing electronic work permits for hazardous chemicals based on large-scale model analysis, characterized in that, Includes the following steps: Step 1: Receive and manage electronic work order files, establish a unique identifier for each work order, and track its review status; Step 2: Extract information from the electronic work order file, identify and structure key information fields including work content, work location, risk identification, safety measures and work time; Step 3: Input the structured key information fields into the pre-trained large model compliance rule library for multi-dimensional compliance review, and output the compliance judgment conclusion of the work ticket, i.e. the review result. The large model compliance rule library is constructed by learning from a large number of labeled compliant and non-compliant work ticket samples and relevant regulatory texts. Step 4: Based on the review results of Step 3, if the work ticket is non-compliant, generate targeted rectification suggestions and provide feedback. Then, receive the rectified work ticket and return to Step 3 for verification until the review is passed; if the work ticket is compliant, approve and archive it. Step 5: Based on historical audit and rectification data, statistically analyze high-frequency violation patterns and use the data to continuously optimize the large model compliance rule library.

2. The method for reviewing electronic work permits for hazardous chemicals based on large-scale model analysis according to claim 1, characterized in that: Step two, which involves extracting information from the electronic work order file, specifically includes: Optical character recognition technology is used to convert work tickets in image or PDF format into raw text information; Based on natural language processing technology, predefined entity fields are identified and extracted from the original text information. The entity fields include at least: work type, hot work level, equipment number, risk point description and safety measure items. The extracted unstructured entity field values ​​are mapped and filled into the corresponding structured fields according to the preset hazardous chemical work order data model, generating standardized work order data objects.

3. The method for reviewing electronic work permits for hazardous chemicals based on large-scale model analysis according to claim 2, characterized in that: Step three, the multi-dimensional compliance review, is executed by the large-scale model compliance rule library, specifically including the compliance score calculation and judgment process based on the output probability of its embedded large-scale model: Let the set of rules used for auditing be R = {r1, r2, ..., rn}, which contains n predefined auditing rules, each rule r i Associate a predefined confidence weight with a rule threshold; The structured key information fields are input into the large model compliance rule library, which then uses the embedded large model to apply the rules r to each rule in the rule set R. i Make an independent judgment and output a probability value representing the degree of compliance of the rule; The probability values ​​of all rules output in the rule set R are weighted and aggregated based on confidence weights to calculate the comprehensive compliance score of the work ticket; The calculation of the comprehensive compliance score is a process of weighted averaging of the output probabilities of all elements in the rule set R. The overall compliance score is compared with the preset compliance threshold. If the overall compliance score is lower than the compliance threshold, an audit result is generated, which determines that the work order is non-compliant.

4. The method for reviewing electronic work permits for hazardous chemicals based on large-scale model analysis according to claim 3, characterized in that: In step three, if the work order is determined to be non-compliant, further steps are taken to locate the violation and classify its severity: Traverse the rule set to identify all rules whose compliance probability is lower than their corresponding rule threshold and mark them as triggering violations; Based on the preset mapping relationship between rule types and severity levels, the severity level of each triggered violation is determined. Violation severity levels include critical violations, major violations, and general violations. Based on the severity level of all triggered violations, an overall violation score is calculated to quantify the degree of non-compliance of the current work order; The system integrates the triggered violations, their corresponding severity levels, and the overall violation score to generate and output the audit details data for the current work order. This audit details data, as part of the audit results, is used to describe the non-compliance in detail.

5. A method for reviewing electronic work permits for hazardous chemicals based on large-scale model analysis according to claim 4, characterized in that: Step four generates targeted rectification suggestions, specifically including: Based on the audit details, including the overall violation score, the triggering violations and their severity levels, the case with the highest similarity is retrieved from the historical rectification case database. The similarity is calculated based on the matching degree of violation type, severity level and work scenario. The violation information of the current work order is compared with the retrieved cases, and effective rectification measures are extracted from the cases; Based on the requirements of the corresponding rules in the large model compliance rule library, the extracted rectification measures are adaptively modified and supplemented to generate a rectification suggestion report containing specific modified fields, modification basis and operation guidelines; In the rectification suggestion report, the violations are sorted and marked according to their severity level. Items identified as critical violations will be prioritized and highlighted, and targeted rectification suggestions will be generated in step four. These suggestions specifically include: Based on the audit details, including the overall violation score, the triggering violations and their severity levels, the case with the highest similarity is retrieved from the historical rectification case database. The similarity is calculated based on the matching degree of violation type, severity level and work scenario. The violation information of the current work order is compared with the retrieved cases, and effective rectification measures are extracted from the cases; Based on the requirements of the corresponding rules in the large model compliance rule library, the extracted rectification measures are adaptively modified and supplemented to generate a rectification suggestion report containing specific modified fields, modification basis and operation guidelines; In the rectification suggestion report, the suggestions are sorted and marked according to the severity level of the violations that triggered them. The suggestions corresponding to the items that are identified as key violations will be prioritized and highlighted.

6. A method for reviewing electronic work permits for hazardous chemicals based on large-scale model analysis according to claim 5, characterized in that: Step four involves verifying the rectified work orders, including a quantitative assessment of the rectification effectiveness: The key information fields of the rectified work order are re-entered into the large model compliance rule library, and the large model embedded in the large model compliance rule library reviews them to obtain a new review result. The result includes a new comprehensive compliance score and a new overall violation score. The new comprehensive compliance score, the new overall violation score, and the preset violation score conversion coefficient are calculated to obtain the score improvement degree before and after rectification. If the improvement rate of the score is greater than the preset improvement threshold, and the new overall violation score is zero, then the rectification verification is deemed to have passed. Otherwise, the rectification is deemed incomplete, and supplementary rectification suggestions are generated based on the remaining violations.

7. A method for reviewing electronic work permits for hazardous chemicals based on large-scale model analysis according to claim 6, characterized in that: Step five involves continuously optimizing the large model compliance rule base, including dynamically updating the confidence weights of its embedded large models: Record the correspondence between the compliance probabilities of each rule output by the large model embedded in the large model compliance rule library and the final confirmed actual compliance status in each audit. The actual compliance status is confirmed by manual review or by verifying the work tickets after rectification. For each rule r i Based on the correspondence, its recent prediction accuracy is calculated periodically; The confidence weights are adjusted based on the prediction accuracy.

8. A method for reviewing electronic work permits for hazardous chemicals based on large-scale model analysis according to claim 7, characterized in that: Step five involves statistical analysis of high-frequency violation patterns, specifically including: Regularly perform cluster analysis on the audit details of historical non-compliant work tickets to identify frequently occurring combinations of triggering violations and form high-risk violation patterns; Calculate the frequency of occurrence and the average overall violation score for each high-risk violation pattern within the statistical period; High-risk violation patterns that occur more frequently than the first preset threshold and whose average overall violation score is higher than the second preset threshold, along with their corresponding work scenario information, are summarized and output to form a risk analysis report.

9. A hazardous chemicals electronic work permit review system based on large-scale model analysis, wherein the system is applied in the review method of any one of claims 1-8, characterized in that: The system includes: The work ticket management module is used to receive and manage electronic work ticket files, establish a unique identifier for each work ticket, and track its review status. The information extraction and structuring module is used to extract information from electronic work order documents, identify and structure key information fields including work content, work location, risk identification, safety measures and work time; The intelligent compliance audit module integrates a pre-trained large-scale model compliance rule library, which is used to receive key information fields and perform multi-dimensional compliance audits. The large-scale model compliance rule library is built by learning from a large number of labeled compliant and non-compliant work ticket samples and relevant regulatory texts, and has a large model embedded in it for performing rule reasoning and judgment. The rectification and verification closed-loop module is used to generate targeted rectification suggestions and provide feedback if the work ticket does not comply with the rules based on the audit results of the intelligent compliance audit module. Then, it receives the rectified work ticket and triggers the intelligent compliance audit module to re-verify until the audit is passed; if the work ticket complies with the rules, it triggers the approval and archiving operation. The data analysis and optimization module is used to statistically analyze high-frequency violation patterns based on historical audit and rectification data, and to continuously optimize the large model compliance rule library using data.

Images