Personnel review method, electronic device, storage medium, and program product

Abstract

The application provides a personnel auditing method, an electronic device, a storage medium and a program product, relates to the technical field of computers, and is used for determining whether an entry material set submitted by a user end is complete, so as to determine whether to issue a work instruction to the user end. The method comprises the following steps: obtaining a work request sent by the user end, the work request comprising a work type identifier and the entry material set; based on the work type identifier, searching a knowledge base to obtain a material list matched with a target work type indicated by the work type identifier; the material list represents qualification information required to be provided by the user; analyzing the entry material set to obtain an analysis result, and comparing the analysis result with the material list to obtain a comparison result; if the entry material set is incomplete, sending a material supplement message to the user end, the material supplement message comprising a missing entry material name; if the entry material set is complete, searching a corresponding work scheme based on the work type identifier, generating a work instruction, and issuing the work instruction to the user end.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a personnel review method, electronic device, storage medium, and program product. Background Technology

[0002] With the rapid development of my country's natural gas transmission industry and the deepening of the national energy strategy, the construction and operation of long-distance natural gas pipelines increasingly rely on professional contractors. However, natural gas transmission is characterized by high pressure, flammability, and explosiveness, posing extremely high requirements for operational safety. Currently, the safety awareness and skill levels of contractors vary widely, resulting in a high proportion of pipeline facility accidents caused by contractors, posing a serious challenge to safety management.

[0003] Traditional contractor management primarily relies on manual on-site inspections and supervision, identifying unsafe conditions through a "person-to-person" approach. This model has significant drawbacks: firstly, massive amounts of data cannot be effectively utilized, and frontline managers are overwhelmed with tedious tasks, making it difficult to focus on core business; secondly, passive and outdated supervision methods are no longer suitable for the needs of the big data and artificial intelligence era, resulting not only in low management efficiency but also contradicting modern proactive and intelligent safety management concepts. Therefore, it is urgent to introduce intelligent technologies to reconstruct the contractor's entire process management system to address the current pain points and bottlenecks in safety management. Summary of the Invention

[0004] This application provides a personnel review method, electronic device, storage medium, and program product. It determines the list of materials that the user needs to provide by identifying the job type, and then confirms whether the set of entry materials provided by the user is complete based on the list of materials, thereby determining whether the user needs to supplement materials or directly issuing a work instruction to the user.

[0005] Firstly, this application provides a personnel review method, comprising: obtaining a job request sent by a user terminal, the job request including a job type identifier and an entry data set; the entry data set including the job qualification information of the user to whom the user terminal belongs; based on the job type identifier, retrieving a list of documents matching the target job type indicated by the job type identifier from a knowledge base; the list of documents representing the qualification information that the user to whom the user terminal belongs needs to provide; parsing the entry data set to obtain a parsing result, and comparing the parsing result with the list of documents to obtain a comparison result; the parsing result representing the entry data set corresponding to the entry data set, and the comparison result representing whether the entry data set is complete; if the comparison result represents that the entry data set is incomplete, sending a data supplement message to the user terminal, the data supplement message including the names of the missing entry documents; if the comparison result represents that the entry data set is complete, retrieving the corresponding job plan from the knowledge base based on the job type identifier, generating a job instruction manual based on the job plan, and issuing the job instruction manual to the user terminal.

[0006] The technical solution provided in this application offers at least the following benefits: It automatically retrieves matching data lists based on work type identifiers; subsequently, it parses the incoming data and automatically compares it with the list, not only identifying data completeness but also triggering supplementary guidance when data is missing, avoiding the inefficiency caused by repeated manual communication. Most importantly, the technical solution directly uses the review results as the trigger condition for generating work instructions, only calling the work plan from the knowledge base and generating customized instructions when qualifications are compliant. This makes qualification review no longer an isolated administrative process but a prerequisite for safe work execution, effectively eliminating the risk of unqualified or incomplete personnel operating illegally, while ensuring that the issued work instructions are highly matched with personnel qualifications and work types, improving the safety and standardization of on-site operations.

[0007] In one possible implementation, the job request further includes a job start and end time, the entry data set includes qualification certificates, and the step of parsing the entry data set to obtain a parsing result, and comparing the parsing result with the data list to obtain a comparison result, includes: The qualification certificate is parsed to obtain the qualification type and the validity period of the qualification certificate; based on the job type identifier, the target qualification type corresponding to the job type is retrieved in the knowledge base to determine whether the qualification type is consistent with the target qualification type; it is determined whether the start and end times of the job both fall within the validity period; if the qualification type is consistent with the target qualification type and the start and end times of the job both fall within the validity period, a comparison result indicating that the entry data set is complete is generated; if the qualification type is inconsistent with the target qualification type, or the start and end times of the job do not both fall within the validity period, a comparison result indicating that the entry data set is incomplete is generated.

[0008] Based on the above implementation method, not only is the type of qualification certificate analyzed, but its validity period is also extracted and compared with the specific start and end times of the work. This ensures that personnel qualifications not only match in category but also fully cover the work cycle in terms of time, effectively eliminating the safety hazards of working with expired certificates or qualifications becoming invalid during the work period. By making the consistency of qualification type and validity of time a necessary and sufficient condition for generating complete data results, a leap from formal review to substantive compliance is achieved. This not only eliminates the risk of oversight in manually verifying validity periods but also blocks violations caused by qualification expiration issues at the source, ensuring the continuous legality and compliance of personnel capabilities throughout the entire work cycle and reducing the probability of safety accidents caused by personnel qualification deficiencies.

[0009] In one possible implementation, parsing the qualification certificate to obtain the qualification type and the validity period of the qualification certificate includes: extracting the text content of the qualification certificate and the location information of the text content on the qualification certificate; comparing whether the location information is consistent with preset location information; the preset location information represents verified standard location information; if consistent, determining the qualification type and the validity period of the qualification certificate based on the text content; if inconsistent, sending a prompt message indicating that the qualification certificate parsing failed to the user terminal.

[0010] Based on the above implementation method, by verifying the positional information of the content text, the reliability and anti-counterfeiting capabilities of qualification certificate parsing are significantly improved. Traditional OCR technology only focuses on text recognition and is easily misled by forged documents or non-standard scanned copies; while this solution compares the physical position of key information (such as validity period and qualification type) on the certificate with the preset standard format before extracting the text. Only when the format structure conforms to the standard is the extracted text data accepted; once the position deviates, the parsing is directly judged as failed and the subsequent process is blocked. This not only effectively filters out fake certificates with altered key information positions, but also avoids data misreading caused by incorrect certificate versions or non-standard scanning, ensuring the authenticity and accuracy of the parsing results of the incoming materials from the data source, laying a reliable data foundation for subsequent qualification comparison.

[0011] In one possible implementation, generating a work instruction based on the work plan includes: identifying semantic information of the work plan based on a large language model, and splitting the work plan into multiple work steps based on the semantic information; the semantic information includes the execution actions of the work plan and the logical relationships between the multiple execution actions; retrieving each work step from the knowledge base and determining the risk event corresponding to each work step; and generating the work instruction based on each work step and the corresponding risk event.

[0012] Based on the above implementation method, by identifying the execution actions and logical relationships in the work plan, unstructured plans can be broken down into ordered work steps, overcoming the shortcomings of traditional template-based instructions, such as poor flexibility and weak targeting. More importantly, this application's solution retrieves associated risk events from the knowledge base for each specific step, directly embedding risk warnings into the work process. This achieves granular alignment between safety measures and work actions. This not only improves the operability and safety of work instructions but also effectively solves the pain point of missing or delayed updates of risk points when manually writing instructions.

[0013] In one possible implementation, the job request further includes a job environment. Determining the risk event corresponding to each job step includes: loading a preset risk map, which includes job nodes, environmental feature nodes, and risk nodes, with the job nodes connected to the environmental feature nodes and the risk nodes; for a target job step, mapping the target job step to a target job node in the preset risk map, mapping the job environment to a target environmental feature node in the preset risk map, and determining the activation probability of each risk node based on the connection relationship between the target job node and each risk node; the target job step is any one of the plurality of job steps, the activation probability represents the probability of occurrence of the risk event corresponding to the risk node, and the connection relationship is related to the job environment; and determining the target risk event corresponding to the target risk node with the highest activation probability when executing the target job step.

[0014] Based on the above implementation method, a pre-defined risk map containing work nodes, environmental feature nodes, and risk nodes is constructed, with the specific work environment as a key variable input. Its core logic lies not only in mapping work steps but also in mapping environmental features as map nodes, using the connections between nodes to calculate the activation probability of each risk node under a specific environment. This derivation mechanism can quantify the weighted impact of environmental factors on the probability of risk occurrence in a specific work step, thereby accurately identifying the risk event with the highest activation probability in the current scenario. This not only solves the pain point of traditional instruction manuals lacking specificity in risk warnings but also achieves differentiated risk early warning based on real-time environmental perception, ensuring that operators can prioritize and prevent the most likely risks in the current environment, significantly improving the accuracy and effectiveness of safety measures.

[0015] In one possible implementation, determining the activation probability of each risk node based on the connection relationship between the target task node and each risk node includes: taking the target task node as the root node, traversing each risk node in the preset risk map, and determining the distance between the target task node and each risk node; determining the risk coefficient of each risk node based on the distance between the target task node and each risk node, wherein the risk coefficient is negatively correlated with the distance; for the target task node, determining the activation probability of each risk node based on the risk coefficient and risk value corresponding to each risk node; wherein the risk value of the risk node is determined based on the risk event corresponding to the risk node.

[0016] Based on the above implementation, the connection distance between work nodes and risk nodes in the graph is transformed into a calculable risk coefficient. Using the negative correlation logic of distance, the degree of closeness between work steps and potential risks is reflected. Furthermore, the inherent risk value of the risk event itself is combined for comprehensive calculation to dynamically derive the activation probability. This implementation not only considers the severity of the risk but also more accurately describes the correlation strength between a specific work step and the risk in the graph structure, effectively avoiding misjudgments caused by the complexity of the graph path. This makes risk warning no longer a simple binary judgment but a continuous probability distribution based on graph theory algorithms, capable of intelligently identifying high-probability, high-hazard risk points most easily triggered under the current work path.

[0017] In one possible implementation, after generating the work instruction based on each work step and the corresponding risk event, the method includes: determining the risk level of the risk event corresponding to each work step, and performing a weighted calculation based on the risk level of the risk event corresponding to each work step to determine the target risk level of the work plan; if the target risk level is greater than a preset risk level, retrieving safety measures corresponding to the target risk level from the knowledge base; and adding the safety measures to the work instruction.

[0018] Based on the above implementation method, the overall hazard level of the work plan is quantified by weighting the risk levels of each step. Once the calculated overall risk exceeds the preset standard, manual judgment is no longer relied upon; instead, a knowledge base retrieval mechanism is automatically triggered to extract and inject high-level safety measures. This makes the work instructions dynamically generated, with more stringent additional protective instructions for higher risks. This not only effectively prevents accidents caused by insufficient protective measures in high-risk operations but also solves the problem of easily overlooking the cumulative effect of overall risk when manually preparing plans.

[0019] Secondly, embodiments of this application provide a personnel verification device, comprising: The acquisition module is used to acquire job requests sent by the user terminal. The job request includes a job type identifier and an entry data set. The entry data set includes the job qualification information of the user to which the user terminal belongs. The processing module is used to retrieve a list of materials matching the target job type indicated by the job type identifier from the knowledge base based on the job type identifier; the list of materials represents the qualification information that the user to which the user terminal belongs needs to provide; The processing module is used to parse the entry data set to obtain the parsing result, and compare the parsing result with the data list to obtain the comparison result; the parsing result represents the entry data list corresponding to the entry data set, and the comparison result represents whether the entry data set is complete; The sending module is configured to send a data supplement message to the user terminal when the comparison result indicates that the entry data set is incomplete. The data supplement message includes the name of the missing entry data. The sending module is used to retrieve the corresponding job plan from the knowledge base based on the job type identifier when the comparison result indicates that the entry data set is complete, generate a job instruction based on the job plan, and send the job instruction to the user terminal.

[0020] In one possible implementation, the processing module is further configured to parse the qualification certificate to obtain the qualification type and the validity period of the qualification certificate; based on the job type identifier, retrieve the target qualification type corresponding to the job type in the knowledge base, and determine whether the qualification type is consistent with the target qualification type; determine whether the start and end times of the job both fall within the validity period; if the qualification type is consistent with the target qualification type and the start and end times of the job both fall within the validity period, then generate a comparison result indicating that the entry data set is complete; if the qualification type is inconsistent with the target qualification type, or the start and end times of the job do not both fall within the validity period, then generate a comparison result indicating that the entry data set is incomplete.

[0021] In one possible implementation, the processing module is further configured to extract the text content from the qualification certificate and the location information of the text content on the qualification certificate; compare the location information with preset location information to see if they are consistent; the preset location information represents verified standard location information; if they are consistent, determine the qualification type and the validity period of the qualification certificate based on the text content; If there is a discrepancy, a message indicating that the qualification certificate resolution failed will be sent to the user terminal.

[0022] In one possible implementation, the sending module is further configured to identify the semantic information of the job scheme based on a large language model, and to split the job scheme into multiple job steps based on the semantic information; the semantic information includes the execution actions of the job scheme and the logical relationships between the multiple execution actions; to retrieve each job step in the knowledge base and determine the risk event corresponding to each job step; and to generate the job instruction based on each job step and the corresponding risk event.

[0023] In one possible implementation, the sending module is further configured to load a preset risk map, which includes job nodes, environmental feature nodes, and risk nodes, wherein the job nodes are connected to the environmental feature nodes and the risk nodes; for a target job step, the target job step is mapped to a target job node in the preset risk map, and the job environment is mapped to a target environmental feature node in the preset risk map; and based on the connection relationship between the target job node and each risk node, the activation probability of each risk node is determined; the target job step is any one of the plurality of job steps, the activation probability represents the probability of occurrence of the risk event corresponding to the risk node, and the connection relationship is related to the job environment; and the target risk event corresponding to the target risk node with the highest activation probability is determined when the target job step is executed.

[0024] In one possible implementation, the sending module is further configured to, with the target job node as the root node, traverse each risk node in the preset risk map to determine the distance between the target job node and each risk node; based on the distance between the target job node and each risk node, determine the risk coefficient of each risk node, wherein the risk coefficient is negatively correlated with the distance; for the target job node, determine the activation probability of each risk node based on the risk coefficient and risk value corresponding to each risk node; the risk value of the risk node is determined based on the risk event corresponding to the risk node.

[0025] In one possible implementation, the processing module is further configured to determine the risk level of the risk event corresponding to each of the work steps, and perform a weighted calculation based on the risk level of the risk event corresponding to each of the work steps to determine the target risk level of the work plan; if the target risk level is greater than a preset risk level, retrieve the safety measures corresponding to the target risk level from the knowledge base; and add the safety measures to the work instruction.

[0026] Thirdly, this application provides an electronic device, including: a processor and a memory, the processor being coupled to the memory; the memory being used to store computer instructions, which are loaded and executed by the processor to enable the computer device to implement the method described in the first aspect.

[0027] Fourthly, this application provides a computer-readable storage medium comprising: computer software instructions; which, when executed in an electronic device, cause the electronic device to implement the method described in the first aspect.

[0028] Fifthly, this application provides a computer program product comprising a computer program; when the computer program is run in an electronic device, it causes the electronic device to implement the method described in the first aspect.

[0029] The beneficial effects of the second to fifth aspects mentioned above are described in the corresponding description of the first aspect and will not be repeated here. Attached Figure Description

[0030] Figure 1 A flowchart illustrating a personnel review method provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of a personnel verification device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0031] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0032] The terms "first" and "second," etc., used in the specification and drawings of this application are used to distinguish different objects or to distinguish different treatments of the same object, rather than to describe a specific order of objects.

[0033] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0034] It should be noted that in the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0035] To facilitate a clear description of the technical solutions of the embodiments of this application, the terms "first" and "second" are used in the embodiments of this application to distinguish the same or similar items with essentially the same function and effect. Those skilled in the art can understand that the terms "first" and "second" are not intended to limit the quantity or execution order.

[0036] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0037] Currently, the traditional contractor operation management model has the following problems: First, the access review mechanism is not strict, and there are loopholes in the source control.

[0038] The selection of contractor teams is the first line of defense for safe operation. However, the current access review process is not rigorous enough, resulting in some contractors' professional qualifications not matching the on-site safety requirements. Due to factors such as personnel shortages, some teams with weak safety awareness and insufficient skills have been able to enter the construction site. This directly leads to the failure to implement the main responsibility for safe production, leaving individual safety responsibilities unfulfilled, and consequently causing repeated violations such as illegal operations, illegal command, and breaches of labor discipline. Such hidden dangers not only threaten daily operational safety but also pose a serious risk to the long-term stable operation of natural gas pipelines.

[0039] Second, safety education is merely a formality, and the effectiveness of training urgently needs to be improved.

[0040] There are significant shortcomings in the current safety education and training for contractors: Rigid approach: Training is mostly rote learning, lacking specificity and interactivity, and fails to make full use of typical accident cases in the natural gas transmission industry for warning education, and historical experience has not been effectively transformed into preventive awareness.

[0041] Insufficient time: Reduced training time prevents contractors from fully digesting and mastering key safety knowledge.

[0042] The assessment is too lenient: the assessment mechanism is not strict enough, so the safety knowledge fails to play a real warning role.

[0043] Third, on-site supervision was neglected, and process control was rendered ineffective.

[0044] In terms of on-site operation supervision, the problem of management deficiencies is particularly prominent: Inadequate implementation of responsibilities: Some on-site management personnel lacked a sense of responsibility, neglected safety supervision, and failed to effectively fulfill their safety assurance duties.

[0045] Poor implementation of measures: The safety control measures formulated often remain on paper and fail to be effectively implemented on site; there have even been serious violations such as starting work without supervision, indicating that the company has not yet formed an effective closed-loop supervision mechanism.

[0046] Frequent violations: Due to a lack of necessary supervision, some contractors prioritize speed over safety, take chances, fail to issue necessary entry permits, and habitually violate regulations despite repeated warnings.

[0047] Lack of emergency response mechanism: The lack of a regular on-site safety audit mechanism makes it impossible to effectively deal with emergencies in a timely manner, creating hidden dangers for accidents.

[0048] Fourth, there are difficulties in utilizing historical data and insufficient digital empowerment.

[0049] Traditional file storage methods severely restrict the extraction of data value: Low retrieval efficiency: The massive amount of unstructured files makes it time-consuming and labor-intensive to find specific information, which greatly reduces work efficiency.

[0050] Information sharing is hindered: Dispersed storage methods impede the rapid flow and sharing of information across departments and regions.

[0051] Lack of decision support: Due to the lack of systematic data analysis, it is difficult to use historical operational data to assess the contractor's long-term performance or identify potential risks, and thus cannot provide strong data support for optimizing future decisions.

[0052] Based on the above problems, this application provides a personnel review method, which determines the list of materials that the user needs to provide by identifying the job type, and then confirms whether the set of entry materials provided by the user is complete based on the list of materials, thereby determining whether the user needs to supplement materials or directly issuing a job instruction to the user.

[0053] The embodiments provided in this application will now be described in detail with reference to the accompanying drawings.

[0054] The personnel verification and matching method provided in this application can be applied to electronic devices. Electronic devices can be servers, such as a server cluster consisting of multiple servers, a single server, a computer, or a processor or processing chip within a server or computer. For example, electronic devices can also be terminals, such as mobile phones, tablets, wearable devices, in-vehicle devices, laptops, ultra-mobile personal computers (UMPCs), netbooks, and personal digital assistants (PDAs). It should be noted that the embodiments of this application do not limit the specific device form of the electronic device.

[0055] This application also provides a personnel review system based on intelligent agents, which is deployed in the aforementioned electronic device and includes a document query module, a personnel qualification review module, and a work instruction generation module.

[0056] The document search module, as the initial entry point and core functional unit of the system, mainly includes an automated document processing engine, which can perform tasks such as compressed file parsing, automatic identification of multi-format document content, and extraction of key information.

[0057] The personnel qualification review module performs structured analysis and intelligent judgment on the information of personnel participating in the operation.

[0058] The work instruction generation module integrates the information from the aforementioned modules to generate an instruction manual that serves to provide guidance and summarize information.

[0059] Figure 1 This is a flowchart illustrating a personnel review method provided in an embodiment of this application. Figure 1 As shown, the personnel review method provided in this application can be implemented using the aforementioned electronic devices.

[0060] The method specifically includes S101 to S105.

[0061] S101. Obtain the job request sent by the user terminal. The job request includes the job type identifier and the set of entry data.

[0062] The entry data set includes the operational qualification information of the user to which the user belongs.

[0063] The prerequisite for reviewing operators is receiving a job request from the user, along with a job type identifier and a set of entry documents sent with it. This information not only clarifies the core context of who applied for the job and in what scenario, but also ensures the uniqueness and traceability of the review object, and transforms unstructured data into a standard data stream that the system can process, providing input for subsequent matching of differentiated standards.

[0064] Specifically, the user terminal refers to the interactive terminal used by contractors or workers, such as mobile applications, web management portals, or handheld smart devices; the work type identifier is a code or label used to distinguish the type of work (such as hot work, high-altitude work), and different identifiers correspond to different safety regulations; the entry document set is a data package consisting of all the supporting documents uploaded by the user, including ID card, special operation certificate, medical examination report, insurance policy, etc.; the work qualification information refers to the key attributes extracted from the documents, such as certificate type, validity period, skill level, etc.

[0065] In one implementation, the electronic device can receive file streams and metadata sent by the user client through an application programming interface (API) built on the backend. Before receiving the data sent by the user client, an interceptor can verify the identity carried by the user client to ensure that the requester is legitimate and has the right to apply. The frontend of the electronic device packages the job type code and the file binary stream and sends them. After receiving the data, the backend first uploads the file to object storage (such as a cloud server), generates a temporary Uniform Resource Locator (URL), and writes the metadata (user ID, job type, and file URL list) into a message queue or directly stores it in the pending status table of the database.

[0066] S102. Based on the job type identifier, retrieve a list of materials in the knowledge base that match the target job type indicated by the job type identifier.

[0067] The data list represents the qualification information that the user on the client side needs to provide.

[0068] Because different types of work have vastly different risks (e.g., hot work requires a welder's certificate, and electrical work requires an electrician's certificate), electronic equipment must retrieve the corresponding access standards from the knowledge base in real time based on the specific work type identifier to ensure that the review criteria match the current work risk. At the same time, business rules and code logic are decoupled so that when regulations or company policies change, only the knowledge base needs to be updated without refactoring the internal logic.

[0069] Specifically, the knowledge base refers to a structured database or rule engine configuration center that stores the mapping relationship between job types and required qualifications; the data list refers to a list of required fields defined for a specific job type, including structured rules such as file type, validity period threshold, and certification authority requirements; and the target job type is a specific business scenario classification parsed from the identifier in the request.

[0070] In one implementation, the electronic device extracts the job type identifier obtained in step S101 and uses it as a query keyword. Next, the electronic device searches the knowledge base to locate a list of documents matching the identifier. This list can be stored in the knowledge base as a configuration file. This list details the required documents for that type of job, the validity period requirements for each type of document, and specific format specifications. To improve response speed, the electronic device prioritizes reading these commonly used rules from a cache; if not found in the cache, it directly queries the knowledge base. Finally, the electronic device organizes the extracted list of documents into a structured list, which serves as the benchmark for subsequent comparisons.

[0071] S103. Analyze the set of incoming data to obtain the analysis results, and compare the analysis results with the data list to obtain the comparison results.

[0072] Among them, the analysis result represents the list of entry data corresponding to the entry data set, and the comparison result represents whether the entry data set is complete.

[0073] Since most of the materials uploaded by users are in image format, PDF format, or compressed file format, the machine cannot directly understand them. It is necessary to use parsing technology to convert them into readable structured information, and then compare them item by item with the data list to quickly identify missing materials or format errors.

[0074] Specifically, the parsing result can be a structured data object (such as document name, validity period, etc.) extracted from the original document using optical character recognition (OCR) and natural language processing technologies; the comparison result can be a Boolean status value or a difference report, indicating whether the submitted materials cover all the required fields in the standard list. The entry document list refers to the list of documents formed after the actual materials submitted by the user have been digitally extracted.

[0075] In one implementation, the electronic device invokes an intelligent recognition service (such as using a large model or based on the device's own recognition capabilities) to extract text content from each image or multi-page document uploaded by the user. After recognition, the electronic device uses text analysis algorithms to locate and extract key information from the recognized text, such as the specific name, number, issuance date, and expiration date of the document, and organizes it into standardized data entries.

[0076] The electronic device then checks each submitted data item against the document list. The check process includes verifying whether any required documents are missing, whether document names match, and whether documents are valid. If all required items are present and meet the time requirements, the data is considered complete; otherwise, the missing or non-compliant items are recorded, and a discrepancy report is generated.

[0077] In one implementation, the job request also includes the job start and end times, and the set of entry materials includes qualification certificates. S103 can be specifically implemented through S1031-S1035: S1031. Analyze the qualification certificate to obtain the qualification type and the validity period of the qualification certificate.

[0078] It is not enough to simply know that a user has uploaded a certificate; it is essential to know exactly what type of certificate it is (e.g., welder's certificate or electrician's certificate) and when it expires.

[0079] Among them, the qualification certificate refers to the legal document issued by the relevant national department or industry association, which proves that the operator has specific skills or qualifications; the qualification type refers to the specific operation category indicated on the certificate, such as "fusion welding and thermal cutting operation", "high-altitude installation, maintenance and dismantling operation", etc.; the validity period refers to the time range of the certificate's legal validity, which usually includes the date of issuance and the date of renewal / expiration.

[0080] In one implementation, the electronic device can call an Optical Character Recognition (OCR) module to scan the certificate image uploaded by the user and extract all text information from the image. Then, using Natural Language Processing (NLP) technology or preset regular expression rules, key fields are located in the extracted text. For example, the electronic device will look for areas containing phrases like "job category" or "permitted project," marking the following text as the qualification type; it will also look for phrases like "validity period" or "until [date]," converting them into a standard time format as the validity period.

[0081] In another implementation, the set of entry documents may also include one or more of the following: identification documents, health, safety and environmental management system (HSE) certification of the guardian, HSE training certificates of the guardian and the person in charge, insurance information, medical examination information, proof of no violations, construction site access permit, quality, health, safety and environment (QHSE) assessment access, QHSE commitment letter, and safety management quiz / internal company quiz results.

[0082] It should be noted that in actual operations, multiple job type identifiers may coexist. In such cases, the electronic device can first retrieve the corresponding data list for each job type identifier from the knowledge base. Then, it takes the union of these data lists to determine the data required from the operator. For all the above materials, the following steps S10311-S10314 are used for content extraction and verification. For some materials, such as answer information, insurance information, and medical examination information, a pass / fail result may also be included. For example, if the medical examination or answer question fails, the operator's data is considered incomplete.

[0083] The QHSE (Quality, Health, and Safety) access assessment system ensures the accuracy and compliance of contractor capability assessments before construction operations through workflow construction and large-scale model analysis technology. The system mainly consists of two parts: loading and verifying the "Contractor Capability Assessment Form for Pre-Construction Operations," and checking the QHSE commitment letter (including the Ten Prohibitions Commitment Letter). First, the system automatically loads the "Contractor Capability Assessment Form for Pre-Construction Operations" from the "QHSE Access Assessment" folder. Next, a large-scale language model is used to comprehensively analyze the table content to determine if the project name is correct, whether the construction unit information meets the requirements, and especially to check the validity period of the qualifications and the matching degree of the business scope. This not only ensures the completeness of the document fields but also intelligently identifies and prompts any non-compliant or risky information, providing accurate data support for subsequent audits. For the QHSE commitment letter section, the relevant files located in the "QHSE Commitment Letter - Ten Prohibitions Commitment Letter" folder are read first. For the "QHSE Commitment Letter," the focus is on checking the construction team's name (which must be stamped for confirmation) and the signature of the project's main responsible person. For the "Ten Prohibitions for Safe Production Commitment Letter," the emphasis is on verifying the signatures of the main responsible person and supervisor. These two steps aim to ensure that all necessary signing and confirmation procedures are completed, thereby meeting the requirements of safety management and regulatory compliance. The entire process utilizes an intelligent agent platform for efficient workflow construction and fully leverages the powerful analytical capabilities of large-scale models to process and verify document content, ensuring that every aspect of the QHSE access assessment system meets the highest standards and requirements. This technical solution not only improves review efficiency but also enhances the accuracy and reliability of data processing, providing a solid foundation for enterprises to implement comprehensive safety management.

[0084] Supported by an intelligent agent platform, the automated safety knowledge assessment system ensures effective verification of the answers given by safety management (guardian) personnel and internal company safety training quizzes through the construction of an efficient workflow and the utilization of large-scale model analysis technology. This system aims to guarantee that the safety knowledge level of all relevant personnel meets enterprise and industry standards. The specific process is as follows: First, electronic devices automatically read the answer records of responsible persons and guardians stored in a designated folder. This process covers a comprehensive acquisition of internal company safety training quiz records. Next, it checks whether the number of participants in these records meets the predetermined requirements, ensuring that each key role participates in the necessary safety knowledge assessment. For each acquired answer file, the system further analyzes the type of respondent (e.g., responsible person, guardian, or ordinary employee) and judges whether their answers meet the requirements based on preset standards. This step not only verifies the accuracy of the participant's identity but also performs in-depth analysis of the answer content through a large-scale language model to confirm whether each participant's mastery of safety knowledge meets the prescribed standards.

[0085] Throughout the process, the intelligent agent platform provides robust workflow support, enabling smooth data processing and analysis. Simultaneously, leveraging the advanced parsing capabilities of large models, it can efficiently and accurately identify and process complex textual information, thereby achieving automated evaluation of responses. This technical solution not only significantly improves review efficiency but also enhances the reliability and impartiality of results, providing strong technical support for enterprise security management. Furthermore, through dynamically generated reports and feedback mechanisms, the system can help enterprises and employees promptly identify and address knowledge gaps, further enhancing overall security awareness and emergency response capabilities.

[0086] Furthermore, with the support of the intelligent agent platform, and by combining the intelligent analysis capabilities of large models with the automated processing capabilities of Python programming, the dynamic generation of standardized work instructions (Word documents) from work plan documents is achieved. Based on the content of the work plan document, a safety work instruction document is automatically generated, including a number, work name, and briefing content (work content, existing hazardous factors and risks, work requirements and precautions).

[0087] First, an empty Word document template needs to be prepared as the basic framework for the final output. This template has all original content removed, retaining only pre-defined placeholders (such as "{Number}", "{Assignment Name}", "{Assignment Content}", etc.) to dynamically populate the generated content later. Next, a Python script is used to read the contents of the assignment plan file and calls a large language model to perform contextual analysis, extracting and generating the required briefing content.

[0088] Specifically, based on the work plan, the work name is first determined; then, the large-scale model analyzes the work content summary in the work plan to generate a detailed description of the work content; next, for the hazard identification and control measures section, the large-scale model further analyzes and generates information on existing hazardous factors and risks; finally, by analyzing relevant content in the work plan, corresponding work requirements and precautions are generated. This generated content is then filled into the corresponding placeholder positions in a Word template according to a preset logical structure. After the placeholders are filled, the Python script writes the final content into a Word document and outputs a work instruction file that conforms to enterprise standards. The entire process relies on the workflow building capabilities of the intelligent agent platform, ensuring that every step from content extraction to document generation is executed efficiently and smoothly. Simultaneously, the intelligent analysis capabilities of the large-scale model ensure that the generated content is both accurate and meets actual needs, avoiding errors or omissions that may occur with manual entry.

[0089] In one implementation, S1031 can be implemented using S10311-S10314: S10311. Extract the text content from the qualification certificate and the location information of the text content on the qualification certificate.

[0090] By simultaneously extracting text and its coordinates on an image, the electronic device obtains not only information about the text itself but also its location. This provides a physical basis for subsequent verification of whether the document format conforms to national or industry standards, ensuring that the parsed data originates from a genuine, standardized document layout, rather than randomly pieced-together text.

[0091] The text content refers to all character sequences identified from the certificate image, including title, name, ID number, job category, date, etc.; the location information is the specific distribution of the text content in the original image coordinate system, usually expressed as the coordinates of the top left and bottom right corners of the rectangle (x1, y1, x2, y2), or the percentage position relative to the length and width of the image.

[0092] In one implementation, the electronic device invokes a recognition engine with layout analysis capabilities. When scanning an image, this engine not only outputs the recognized text strings but also simultaneously outputs the coordinate bounding box data corresponding to each text block or paragraph. The system packages these two parts of data into a structured intermediate object. For example, if a user uploads a welder's certificate, the system recognizes that the four characters for "job category" are located in the coordinate region (50, 100) to (150, 120), and the two characters for "welding" to its right are located in the region (160, 100) to (220, 120). The system extracts these texts along with their corresponding coordinates.

[0093] S10312. Compare the location information with the preset location information to see if they are consistent.

[0094] Among them, the preset location information represents the verified standard location information.

[0095] Generally, legitimate qualification certificates have strict and standardized formatting, and the positions of key fields (such as validity period and job type) are fixed. Counterfeit certificates can often imitate the text content, but it is difficult to perfectly replicate the precise layout (especially after multiple photocopies or Photoshop manipulation, the positions will shift). By comparing the actual extracted positions with the standard template positions, fake or non-standard certificates with abnormal layouts can be quickly identified, preventing misleading information from being extracted based on incorrect layouts.

[0096] Specifically, the preset location information refers to the standard template data stored in the electronic device, which represents the coordinate range or relative positional relationship that the key fields of this type of qualification certificate should have under ideal conditions (or after statistical averaging of a large number of real certificates).

[0097] In one implementation, the system first loads the corresponding standard template library based on the certificate's preliminary classification (e.g., identifying it as a special operations certificate through logo or title recognition). Then, the electronic device geometrically compares the actual coordinates of key fields extracted in the previous steps (such as the validity period label and its corresponding value) with the standard coordinates defined in the template. The comparison algorithm calculates the center point distance, overlap rate, or relative position ratio of the coordinates. If the deviation is less than a preset tolerance threshold (e.g., allowing ±5% scanning deformation error), it is considered consistent; if the deviation is too large (e.g., the validity period field is at the edge of the image or completely misaligned), it is considered inconsistent.

[0098] S10313. If consistent, determine the qualification type and the validity period of the qualification certificate based on the text content.

[0099] After ensuring the authenticity of the document format, the final content verification is conducted. Only after the physical layout verification is passed is the extracted text content considered credible. The data, after content recognition and format verification, is formally transformed into structured information usable by the business (qualification type, validity period) as direct input for subsequent review logic, ensuring the reliability and accuracy of the data source.

[0100] In one implementation, once the location information is confirmed, the electronic device can be certain that the text located within a specific coordinate region is the target content. It directly reads the text string within that coordinate region. For qualification types, the electronic device may also need to perform a simple dictionary match (mapping the identified colloquial description to standard code); for validity periods, the electronic device converts the read date string into a standard time object.

[0101] S10314. If they are inconsistent, send a message to the user indicating that the qualification certificate resolution failed.

[0102] When an abnormal document format is detected (highly likely a fake, altered, or severely damaged scanned copy), the electronic device cannot force its way through; otherwise, it may produce an incorrect verification conclusion. Therefore, the current process must be terminated to prevent the risk from spreading. The user should be clearly informed that the failure is due to a mismatch in document format / layout, rather than simply missing content. The user should be guided to retake a clear, complete, and legitimate photograph of the document, or warned against using counterfeit documents.

[0103] In one implementation, upon detecting a location comparison failure, the system halts subsequent text extraction and logical judgment. The system constructs a specific error response code and generates a corresponding message, such as: "Abnormal certificate layout detected. Please ensure the document is original and photographed correctly. Do not use image editing software." The electronic device then returns this message to the user client via an application programming interface (API), displaying a warning message and rejecting the upload. Simultaneously, the electronic device can record this abnormal behavior in the background (including image fingerprints and deviation data) for subsequent risk control audits or manual review, preventing malicious attempts to crack the system.

[0104] S1032. Based on the job type identifier, retrieve the target qualification type corresponding to the job type in the knowledge base, and determine whether the qualification type is consistent with the target qualification type.

[0105] Different work scenarios require completely different professional skills; an electrician's certificate holder cannot perform welding work, and vice versa. By ensuring that the type of certificate held by the worker corresponds to the job applied for, violations and safety accidents caused by skill mismatches can be prevented.

[0106] In one implementation, the electronic device first reads the job type identifier (e.g., hot work) from the job request, and then queries the knowledge base for the target qualification type corresponding to that identifier (e.g., the knowledge base specifies that hot work must be performed with a "welding and thermal cutting" certificate). Next, the electronic device performs a string comparison or mapping relationship verification between the parsed user qualification type and the found target qualification type.

[0107] S1033. Determine whether the start and end times of the operation fall within the validity period.

[0108] Many safety incidents occur when certificates have just expired or are on the verge of expiration. Therefore, verifying that all work operations are within their validity period can eliminate the risk of unlicensed work or the use of expired certificates, ensuring that personnel qualifications remain compliant throughout the work process.

[0109] The start and end times of the job refer to the planned start and end times of the work applied for by the user, which define a continuous time interval.

[0110] In one implementation, the electronic device obtains the start and end times from the job request, as well as the parsed certificate validity and expiration dates. It then performs two logical checks: first, it checks if the job start time is later than the certificate validity date; second, it checks if the job end time is earlier than the certificate expiration date. Only when both conditions are met is the job considered to be within its validity period.

[0111] S1034. If the qualification type is consistent with the target qualification type, and the start and end times of the operation both fall within the validity period, then a comparison result representing the complete set of entry data will be generated.

[0112] Electronic devices can only be deemed to have a valid qualification certificate if both the type matching and time validity dimensions are met.

[0113] In one implementation, the electronic device can have a logic checker in memory. When it receives a type matching success signal and a time verification success signal from a previous step, the logic checker outputs a True value. The electronic device then constructs a standard success response object, marks the object's audit status field as complete or passed, and marks the certificate as verified. This result is summarized in the overall entry document audit conclusion as the basis for allowing entry to the next stage.

[0114] S1035. If the qualification type is inconsistent with the target qualification type, or if the start and end times of the operation do not both fall within the validity period, a comparison result indicating that the entry data set is incomplete will be generated.

[0115] To ensure safety, the safety audit follows the "weakest link" principle: if any key indicator (type or time) fails to meet the standard, the entire audit is deemed unqualified. This audit logic can promptly block non-compliant work applications and clearly identify the reasons for failure (whether it's due to mismatched certification or expired certification), allowing users to make targeted rectifications.

[0116] In one implementation, if there is a type mismatch, the electronic device will write the error code "TYPE_MISMATCH" into the result object, along with a message such as "Your certificate type does not match the job requirements".

[0117] If the time expires, the electronic device writes the error code "EXPIRED_CERT" and indicates that the certificate has expired during the operation.

[0118] The electronic device encapsulates these error messages into an incomplete comparison result and passes it to the main process. Upon receiving this result, the main process triggers the data supplement message sending logic, informing the user of the specific cause of the non-compliance and requesting a certificate replacement or adjustment of the job time.

[0119] S104. If the comparison results indicate that the entry data set is incomplete, a data supplement message is sent to the user terminal. The data supplement message includes the names of the missing entry data.

[0120] When missing materials are discovered, the process should be stopped immediately and the missing items should be notified. This can prevent users from wasting time by waiting blindly or repeatedly submitting invalid applications, and it can also prevent non-compliant personnel from entering the preparation stage for subsequent high-risk operations.

[0121] The supplementary information message indicates a notification data packet containing clear error guidance; the missing entry information name refers to the specific item name that exists in the standard list but was not provided by the user or failed to be parsed (such as a medical examination certificate).

[0122] In one implementation, if the comparison results show that the data is incomplete, the electronic device can immediately terminate the current review process and extract the specific missing document name or the reason for non-compliance based on the previously generated discrepancy report.

[0123] The electronic device assembles this information into a clear prompt, explicitly indicating which materials the user needs to submit or what content needs to be corrected. The device then returns this prompt directly to the user's interface via an interface, displaying it as a pop-up or highlighted in red, or notifying the user via SMS or in-app push notifications. Simultaneously, the device updates the application status to "Pending Supplementation" and retains the submitted valid data, so that the user doesn't need to re-enter all information after supplementation; they only need to address the missing items.

[0124] S105. If the comparison result indicates that the set of entry data is complete, the corresponding work plan is retrieved from the knowledge base based on the work type identifier, and a work instruction is generated based on the work plan and sent to the user terminal.

[0125] Once the set of entry documents submitted by the user is approved, it means that the electronic equipment has confirmed the compliance of the user's qualifications. At this time, the work instructions can be issued to the user to ensure that the workers have mastered the relevant safety operation requirements before starting work.

[0126] The work plan refers to the standardized work template stored in the knowledge base, which includes the common risk events, standard operating procedures and control measures for this type of work; the work instruction refers to the final executable document (such as PDF or H5 page), which integrates the general plan with the user's specific information (such as name, qualification level, and specific risk warnings).

[0127] In one implementation, after confirming the completeness of the data, the electronic device retrieves the standard construction plan template and safety measure library for that type of work from the knowledge base based on the work type identifier. Then, the electronic device replaces the placeholders in the templates with the current user's real information (such as name, job type, specific work time and location) and key parameters parsed from the qualification documents.

[0128] Electronic devices can dynamically insert specific warnings or operational prohibitions based on the risk level. Once all content is assembled, it can be rendered into an unalterable portable document format (such as PDF) or generated as an interactive web page.

[0129] Finally, the electronic device pushes the access link of the document to the user's terminal, forcing the user to open and read it on the interface, and requiring the user to make an electronic signature or click the confirmation button. Only after this confirmation is completed will the document be officially released, allowing the user to proceed to the subsequent work permit issuance process.

[0130] In one implementation, generating a work instruction based on a work plan can be achieved in the following way: Based on a large language model, the semantic information of the task plan is identified, and the task plan is decomposed based on the semantic information to obtain multiple task steps.

[0131] The semantic information includes the execution actions of the job plan and the logical relationships between multiple execution actions.

[0132] Traditional operational plans are often large blocks of loosely structured natural language text, making it difficult for machines to directly map them to specific risk control points. However, by leveraging the semantic understanding capabilities of large language models, unstructured text can be transformed into structured sequences of actions. By identifying the actions to be executed and their logical relationships (such as sequence, parallel relationships, and conditional branches), electronic devices can break down complex plans into the smallest executable units.

[0133] Specifically, semantic information refers to the deeper meaning contained in the text, which can refer to what to do (perform an action, such as opening a valve or detecting gas) and how / when to do it (logical relationship, such as testing before operation, "stop if concentration exceeds the standard"); operation steps refer to independent, atomized operation instruction units extracted from the plan, which are the basic constituent elements of the instruction manual.

[0134] In one implementation, the electronic device inputs the full text of the retrieved job plan as prompts into a large language model, and pre-sets instructions to require the model to extract all execution actions and analyze their logical sequence. The large model analyzes verbs, conjunctions, and conditional clauses in the text, and outputs a structured list (such as JSON format) containing information such as step numbers, step descriptions, and prerequisite dependencies.

[0135] Retrieve each job step from the knowledge base and identify the risk event corresponding to each job step.

[0136] General work procedures may only describe standard operations but fail to clearly identify the specific risks at each step, or the risk descriptions may not be specific enough. By retrieving specific risk events from a knowledge base for each broken-down step, targeted protective measures can be provided. This ensures that workers are immediately aware of the potential accidents that each action may cause (e.g., opening a valve carries the risk of high-pressure jetting, entering a confined space carries the risk of oxygen deprivation and asphyxiation), thereby improving the relevance and effectiveness of safety warnings.

[0137] In one implementation, the electronic device extracts key feature words (such as gas detection, hot work, and climbing) for each work step generated in step one. Then, the electronic device performs fuzzy matching or vector retrieval in a risk knowledge base to find the risk entry that best matches the action feature. The search results are typically a list containing all the major risks that the step may face. The electronic device then associates this risk information with the original step data, forming a "step + risk" data structure.

[0138] A work instruction manual is generated based on each work step and the corresponding risk event.

[0139] The preceding steps addressed the issues of implementation and protection, respectively. The next step is to integrate the two to generate a comprehensive guidance document that includes both operational procedures and embedded safety warnings.

[0140] In one implementation, the electronic device invokes a document generation module (which could be a template-based rendering engine or a large language model for further refinement). The list of steps and their associated risk events are used as input data. During this process, the content is arranged logically according to the steps. Below each step, a description of the corresponding risk event, specific preventative measures (e.g., wearing protective goggles is mandatory), and emergency response methods are automatically inserted. Finally, the integrated content is converted into PDF, H5 webpages, or interactive cards. The large language model can be used to transform rigid data into easily understandable instructions, such as: "When performing step X, please pay special attention to risk Y and be sure to take measure Z."

[0141] In one implementation, the job request also includes the job environment, identifying the risk events corresponding to each job step, which can be achieved as follows: Load a preset risk map, which includes operation nodes, environmental feature nodes, and risk nodes. The operation nodes are connected to the environmental feature nodes and risk nodes.

[0142] Traditional risk databases are typically flat lists (e.g., actions directly point to corresponding risks), neglecting the crucial variable of environment. The same work action (such as welding) carries drastically different risks in a drying workshop and inside a storage tank filled with flammable gases. By introducing a risk graph data structure, the aim is to construct a multi-dimensional network of connections, dynamically linking personnel actions (work nodes), site conditions (environmental characteristic nodes), and consequential accidents (risk nodes). Loading this graph enables electronic equipment to possess context-aware capabilities, allowing it to dynamically adjust its risk assessment logic based on specific environmental characteristics, rather than rigidly applying general rules.

[0143] The pre-defined risk graph is a graph-based data structure composed of nodes (entities) and edges (relationships). It stores not only data but also the logical connections between entities. Operational nodes represent specific operational actions (e.g., hot work, working at heights, hoisting, etc.). Environmental characteristic nodes represent the physical or chemical state of the work site (e.g., confined space, high temperature, presence of flammable materials, humidity, nighttime, etc.). Risk nodes represent potential accident types (e.g., fire, fall, electric shock, poisoning, etc.). Connection relationships refer to the lines between nodes, representing causal or influence relationships. Specifically, the strength or properties of the connection between operational nodes and risk nodes are constrained by the environmental characteristic nodes.

[0144] In one implementation, when an electronic device starts up or receives a request, it reads predefined graph data from a graph database (such as Neo4j) or a serialized file. This data has been trained or configured in the background by security experts, defining thousands of nodes and their interrelationships. For example, a logical chain already exists in the graph: hot work (work node) connects to fire (risk node), but this connection carries a conditional function that depends on the presence of flammable materials (environmental characteristic node). The loading process instantiates this vast network of relationships into memory, ready to receive queries.

[0145] For the target operation steps, the target operation steps are mapped to target operation nodes in the preset risk map, and the operation environment is mapped to target environment feature nodes in the preset risk map. Based on the connection relationship between the target operation nodes and each risk node, the activation probability of each risk node is determined.

[0146] The target operation step is any one of multiple operation steps, the activation probability represents the probability of the occurrence of the risk event corresponding to the risk node, and the connection relationship is related to the operation environment.

[0147] Because static rules cannot cope with ever-changing on-site environments, electronic devices can use predefined topological structures and weighting algorithms in the graph to calculate the real-time probability (activation probability) of various risks arising from executing the step in the current specific environment by mapping the current target operation step and the specific operation environment to corresponding nodes in the graph. This upgrades risk assessment from qualitative (whether there is a risk) to quantitative (how great the risk is), thereby accurately identifying the most dangerous hidden dangers in the current scenario.

[0148] Specifically, the target operation step refers to the specific operational instruction currently being analyzed; the activation probability is a value (or percentage) between 0 and 1, representing the likelihood of the risk node being triggered under the current environmental conditions; the higher the probability, the more urgent the risk; the connection relationship is related to the environment, meaning that the weight or on / off state of the edges in the graph is dynamic. For example, if the environmental node shows no flammable materials, the connection weight from hot work to fire may approach 0; if it shows high concentration of methane, the weight will approach 1.

[0149] In one implementation, the electronic device first uses a semantic similarity algorithm to map the target operation steps described in the text (such as welding on the top of the storage tank) to welding operation nodes in the graph; and maps the operation environment (such as "residual oil and gas in the storage tank, poor ventilation") to environmental feature nodes such as the presence of flammable gases and confined space.

[0150] Then, after locating these nodes in the map, the electronic equipment traverses the path from the target operation node, through the influence of the target environmental feature nodes, and finally reaches each risk node.

[0151] Finally, the electronic device dynamically calculates the activation probability of risk nodes on each path based on a preset algorithm (such as Bayesian network inference or weighted summation) and the state values ​​of environmental nodes.

[0152] For example, for the welding step, if the environment is mapped to a dry open area, the calculated fire activation probability is 0.05 and the electric shock probability is 0.1; if the environment is mapped to an oil and gas environment, the algorithm will significantly increase the fire / explosion activation probability to 0.95, while also increasing the poisoning probability.

[0153] Given a defined target operation step, identify the target risk event corresponding to the target risk node with the highest activation probability.

[0154] In the context of limited resources and precious attention at work sites, instruction manuals cannot list all theoretically existing risks. Therefore, they need to focus on the most dangerous and likely potential hazards. By selecting the risks with the highest activation probability, the risk warnings are highlighted. This ensures that the generated work instructions can immediately inform workers of the greatest danger in this environment, guiding them to prioritize the most critical prevention and control measures and preventing key risks from being overlooked due to information overload.

[0155] In one implementation, the electronic device, based on the probability list obtained in the previous step (e.g., fire: 0.95, electric shock: 0.1, fall: 0.05), executes a sorting algorithm (such as the Max function) to find the item with the largest value. Then, it iterates through all associated risk node probability values, selects the node with the highest probability value (such as the fire node in the example above), extracts its associated natural language description, consequence level, and recommended core control measures from that node, and marks it as the target risk event.

[0156] In one implementation, the activation probability of each risk node is determined based on the connection relationship between the target job node and each risk node. This can be achieved in the following way: Using the target task node as the root node, traverse each risk node in the preset risk map to determine the distance between the target task node and each risk node.

[0157] In real-world knowledge graphs, the connection between risks and operational actions is often not direct, but rather transmitted through a series of intermediate conditions (such as environmental factors, equipment status, and personnel qualifications). In graph theory, the distance between two nodes (usually referring to the number of edges or weighted length of the shortest path) represents the length of the causal chain. The shorter the distance, the greater the likelihood that the risk is directly caused by the current operation, and the more direct the causal relationship; the longer the distance, the more intermediate conditions need to be met simultaneously to trigger the risk, and the weaker the correlation. Calculating distance is a fundamental geometric indicator for subsequent quantitative risk assessment.

[0158] Here, the root node refers to the starting point of the calculation, i.e., the target task node currently being analyzed. All distance measurements begin here. Distance refers to the minimum number of edges traversed from the target task node to a certain risk node in the graph structure, or the sum of the weights of the traversed edges. It represents the path length of causal transmission.

[0159] In one implementation, the system initializes a queue, adds the target job node to the queue, and marks its distance as 0. Then, a loop begins: a node is retrieved from the queue, and all its neighboring nodes are checked. If a neighboring node is an unvisited risk node or an intermediate node, its distance is marked as the current node's distance + 1 or the current node's distance + edge weight, and it is added to the queue. This process continues until all reachable risk nodes have been visited and the shortest distance has been recorded.

[0160] For example, the target work node is hot work. Path A: Hot work → Flammable materials → Fire. Distance is 2. Path B: Hot work → Unauthorized operation → Lack of supervision → Fire. Distance is 3. The electronic equipment will record the shortest distance between the fire node and the hot work node as 2.

[0161] Based on the distance between the target operation node and each risk node, the risk coefficient of each risk node is determined, and the risk coefficient is negatively correlated with the distance.

[0162] The longer the causal chain, the higher the uncertainty, and the lower the actual probability of occurrence. By establishing negative correlations (the greater the distance, the smaller the coefficient), electronic devices can automatically reduce the weight of indirect risks and highlight direct risks. This makes the assessment model conform to the objective laws in safety engineering, namely that direct hazards are more urgent and have a higher probability of occurrence than indirect hazards.

[0163] The risk coefficient is a normalized value between 0 and 1, representing the credibility or basic weight of risk derived solely from the dimension of the length of the causal path.

[0164] In one implementation, the electronic device can pre-determine a decay function to calculate the risk coefficient based on the distance. For example, the risk coefficient could be the derivative or exponential result of the distance, or it could have a piecewise mapping relationship with the distance. For instance, for a fire node, the distance is 2; if calculated using the reciprocal relationship, the risk coefficient is 1 / 2 = 0.5. For another device fatigue fracture risk that requires an extremely long chain to trigger, the distance is 5, and 1 / 5 = 0.2. Clearly, the fire node has a higher risk coefficient.

[0165] For each target task node, the activation probability of each risk node is determined based on the risk coefficient and risk value corresponding to each risk node. The risk value of a risk node is determined based on the risk event corresponding to the risk node.

[0166] In practical operations, merely assessing the probability trend (coefficient determined by distance) is insufficient; the inherent severity or baseline occurrence rate of the risk event itself (i.e., risk value) must also be considered. Some risks, despite their short paths (high coefficient), are extremely difficult to occur in industrial scenarios (low baseline value); others have longer paths, but their occurrence would be catastrophic (high baseline value). Therefore, integrating path correlation (coefficient) with the inherent risk of the event (risk value) yields a comprehensive and dynamic activation probability. This ensures that the assessment results consider both the directness of the current scenario and the essential attributes of the risk event, achieving precise quantification across multiple dimensions.

[0167] Specifically, the risk value refers to a static or semi-static value stored in the risk node attribute, which represents the basic hazard, historical frequency of occurrence, or severity level of the risk event under normal circumstances (e.g., the risk value of an explosion is usually higher than that of a minor abrasion).

[0168] In one implementation, the electronic device reads two input values ​​for each risk node: a risk coefficient and a risk value. The risk value can be read from the graph node attributes and reflects the inherent magnitude of the risk (which can be preliminarily modified based on environmental characteristics or can be a base value).

[0169] Electronic devices calculate activation probabilities using weighted products or specific algorithm models. Taking weighted products as an example, even if the inherent value of a risk is large, if it is far from the current task, the final activation probability will be lowered; conversely, if it is close and the inherent value is high, the probability will reach its peak.

[0170] For example, risk A (fire) is close by, with a coefficient of 0.8 and an inherent explosion risk value of 0.9. The activation probability is 0.8. 0.9 = 0.72, which is a high probability and requires special attention.

[0171] For example, Risk B (nuclear radiation) has a distance of 0.1, an inherent hazard risk value of 1.0, and an activation probability of 0.1. 1.0 = 0.1, which is a low probability and is unlikely to be triggered in the current work scenario.

[0172] In one implementation, after generating the work instruction, the method further includes the following steps: Determine the risk level of the risk event corresponding to each work step, and perform a weighted calculation based on the risk level of the risk event corresponding to each work step to determine the target risk level of the work plan.

[0173] The safety of a work instruction depends not only on the risk of individual steps but also on the overall risk potential of the entire plan. While a single step may have low risk, the cumulative effect of multiple medium- to high-risk steps, or the presence of a single critical step with extremely high risk, can drastically increase the overall hazard level of the entire work plan. Therefore, quantifying the overall risk level (target risk level) of the entire work plan from a global perspective is crucial. Through weighted calculations, the system can identify plans with a limited number of steps but highly concentrated risks, preventing the underestimation of the overall operational hazard due to a limited local perspective. This provides a quantitative basis for determining whether subsequent control measures need to be upgraded.

[0174] Among them, the risk level refers to the classification of the severity of the risk event (such as: low, medium, high, very high, or numerical level 1-5), which is usually based on a comprehensive assessment of the severity of the accident consequences and the probability of occurrence.

[0175] In one implementation, the electronic device first iterates through each work step in the instruction manual, reads the associated risk events, and converts them into numerical risk levels according to a preset standard table (e.g., level 5 for fatal accidents and level 2 for minor injuries).

[0176] Then, the electronic device assigns weights to each step based on its attributes, such as whether it involves special operations or whether it is on the critical path.

[0177] Finally, the electronic device calculates the target risk level of the entire scheme using a weighted summation / average or aggregation algorithm.

[0178] For example, a certain solution has 3 steps. Step 1 (low risk, level 1, weight 1); Step 2 (medium risk, level 3, weight 2); Step 3 (high risk, level 5, weight 3, because it is the core hot work step).

[0179] Target risk level is calculated using a weighted average: (1×1+3×2+5×3) / (1+2+3)=(1+6+15) / 6=22 / 6≈3.67 The target risk level is approximately 3.7 (close to the high risk level).

[0180] If the target risk level is higher than the preset risk level, retrieve the safety measures corresponding to the target risk level from the knowledge base.

[0181] Standard operating procedures typically include general safety tips, but these tips are often insufficient to prevent accidents from occurring in situations deemed high-risk. Therefore, when the overall risk exceeds a preset safety threshold (preset risk level), additional intervention is triggered. By retrieving enhanced safety measures from a knowledge base that match the high-risk level, a differentiated safety management strategy is implemented where higher risks equate to stricter controls, ensuring that high-risk operations receive adequate and robust protective resources.

[0182] The preset risk level refers to the safety threshold built into the electronic device. Exceeding this value means that the operation falls under the scope of key supervision or special control. Safety measures refer to additional control measures specifically designed for specific high-risk levels, which usually include upgrading the approval process, increasing the number of supervisors, mandating the use of specific advanced protective equipment, and implementing real-time monitoring, which are different from regular safety prompts.

[0183] In one implementation, the electronic device compares the calculated target risk level with a preset risk level. If the target risk level is higher than the preset risk level, the electronic device extracts the risk level range and sends a query request to the knowledge base to find additional safety measures applicable to the high-risk level. It then receives a list of specific measures from the knowledge base, such as: a special-grade hot work permit is required; at least two gas detectors must be used; a dedicated safety officer must be present throughout the operation; and an emergency drill must be conducted before the operation.

[0184] If the target risk level is not greater than the preset risk level, the search will not be performed, and the process will proceed directly to the next step or end.

[0185] Add safety measures to the work instructions.

[0186] If the retrieved safety measures are not integrated into the instruction manual, they remain merely isolated database records, unable to be communicated to frontline workers. Dynamically inserting these targeted, high-level safety measures into the instruction manual ensures that workers performing high-risk tasks can clearly see and be compelled to comply with these additional provisions. This transforms the work instruction manual from a standard operating manual into a customized high-risk management plan, truly achieving the integration of technical and management measures.

[0187] In one implementation, all additional measures (e.g., "This operation is high-risk, requiring dual supervision") are listed in red font on the cover or first page of the instruction manual. These measures are then linked to specific operational steps. For example, a mandatory checklist for the "Operation in Progress" step is linked to a 30-minute inspection. If a measure involves approval, corresponding electronic approval requirements are automatically generated. These new measures are highlighted, bolded, or have confirmation checklists added, forcing users to read and confirm them before execution. Finally, a new version of the instruction manual is saved and marked as a high-risk enhanced version.

[0188] This application embodiment can divide the following device into functional modules based on the above method example. For example, each function can be divided into its own functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. Optionally, the module division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0189] In some embodiments, this application also provides a personnel verification device. The personnel verification device may include one or more functional modules for implementing the methods described in the above embodiments. For example, Figure 2 This is a schematic diagram of the structure of a personnel verification device 200 provided in an embodiment of this application. Figure 2 As shown, the personnel verification device 200 includes: The acquisition module 210 is used to acquire the job request sent by the user terminal. The job request includes a job type identifier and an entry data set. The entry data set includes the job qualification information of the user to which the user terminal belongs. Processing module 220 is used to retrieve a list of materials matching the target job type indicated by the job type identifier from the knowledge base based on the job type identifier; the list of materials represents the qualification information that the user to which the user belongs needs to provide. The processing module 220 is used to parse the incoming data set, obtain the parsing results, and compare the parsing results with the data list to obtain the comparison results; the parsing results represent the incoming data list corresponding to the incoming data set, and the comparison results represent whether the incoming data set is complete; The sending module 230 is used to send a data supplement message to the user terminal when the comparison result indicates that the entry data set is incomplete. The data supplement message includes the name of the missing entry data. The sending module 230 is used to retrieve the corresponding work plan from the knowledge base based on the work type identifier when the comparison result indicates that the set of incoming data is complete, and to generate a work instruction based on the work plan and send the work instruction to the user terminal.

[0190] In one possible implementation, the processing module 220 is further configured to parse the qualification certificate to obtain the qualification type and the validity period of the qualification certificate; based on the job type identifier, retrieve the target qualification type corresponding to the job type in the knowledge base to determine whether the qualification type is consistent with the target qualification type; determine whether the start and end times of the job both fall within the validity period; if the qualification type is consistent with the target qualification type and the start and end times of the job both fall within the validity period, then generate a comparison result indicating that the entry data set is complete; if the qualification type is inconsistent with the target qualification type, or the start and end times of the job do not both fall within the validity period, then generate a comparison result indicating that the entry data set is incomplete.

[0191] In one possible implementation, the processing module 220 is further configured to extract the text content in the qualification certificate and the location information of the text content on the qualification certificate; compare the location information with preset location information; the preset location information represents the verified standard location information; if they match, determine the qualification type and the validity period of the qualification certificate based on the text content; if they do not match, send a prompt message indicating that the qualification certificate parsing failed to the user terminal.

[0192] In one possible implementation, the sending module 230 is further configured to identify the semantic information of the job plan based on a large language model, and to split the job plan into multiple job steps based on the semantic information; the semantic information includes the execution actions of the job plan and the logical relationships between the multiple execution actions; to retrieve each job step in the knowledge base and determine the risk event corresponding to each job step; and to generate a job instruction manual based on each job step and the corresponding risk event.

[0193] In one possible implementation, the sending module 230 is further configured to load a preset risk map, which includes job nodes, environmental feature nodes, and risk nodes, with job nodes connected to environmental feature nodes and risk nodes. For a target job step, the target job step is mapped to a target job node in the preset risk map, and the job environment is mapped to a target environmental feature node in the preset risk map. Based on the connection relationship between the target job node and each risk node, the activation probability of each risk node is determined. The target job step is any one of multiple job steps, and the activation probability represents the probability of occurrence of the risk event corresponding to the risk node. The connection relationship is related to the job environment. The target risk event corresponding to the target risk node with the highest activation probability is determined when the target job step is executed.

[0194] In one possible implementation, the sending module 230 is further configured to traverse each risk node in the preset risk map with the target job node as the root node, and determine the distance between the target job node and each risk node; based on the distance between the target job node and each risk node, determine the risk coefficient of each risk node, wherein the risk coefficient is negatively correlated with the distance; for the target job node, determine the activation probability of each risk node based on the risk coefficient and risk value corresponding to each risk node; the risk value of the risk node is determined based on the risk event corresponding to the risk node.

[0195] In one possible implementation, the processing module 220 is further configured to determine the risk level of the risk event corresponding to each work step, and perform a weighted calculation based on the risk level of the risk event corresponding to each work step to determine the target risk level of the work plan; if the target risk level is greater than the preset risk level, retrieve the safety measures corresponding to the target risk level from the knowledge base; and add the safety measures to the work instruction.

[0196] When the methods of the above embodiments are implemented in hardware, this embodiment of the invention provides a possible structural schematic diagram of the electronic device involved in the above embodiments. For example... Figure 3 As shown, the electronic device 300 includes: a processor 302, a communication interface 303, and a bus 304. Optionally, the electronic device 300 may also include a memory 301.

[0197] Processor 302 may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 302 may be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 302 may also be a combination that implements computing functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0198] Communication interface 303 is used to connect with other devices via a communication network. This communication network can be Ethernet, wireless access network, wireless local area network (WLAN), etc.

[0199] The memory 301 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0200] In one possible implementation, the memory 301 can exist independently of the processor 302. The memory 301 can be connected to the processor 302 via a bus 304 and is used to store instructions or program code. When the processor 302 calls and executes the instructions or program code stored in the memory 301, it can implement the method provided in the embodiments of the present invention.

[0201] In another possible implementation, the memory 301 can also be integrated with the processor 302.

[0202] Bus 304 can be an Extended Industry Standard Architecture (EISA) bus, etc. Bus 304 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by only one line, but this does not mean that there is only one bus or one type of bus.

[0203] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the service calling device can be divided into different functional modules to complete all or part of the functions described above.

[0204] This application also provides a computer-readable storage medium. All or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware. The program can be stored in the aforementioned computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be any of the foregoing embodiments or memory. The aforementioned computer-readable storage medium can also be an external storage device of the aforementioned device, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the aforementioned device. Further, the aforementioned computer-readable storage medium can include both internal storage units of the aforementioned service call device and external storage devices. The aforementioned computer-readable storage medium is used to store the aforementioned computer program and other programs and data required by the aforementioned device. The aforementioned computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0205] This application also provides a computer program product comprising a computer program that, when run on a computer, causes the computer to perform the methods provided in the above embodiments.

[0206] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A personnel review method, characterized in that, The method includes: Obtain a job request sent by the user terminal, the job request including a job type identifier and an entry data set; the entry data set includes the job qualification information of the user to whom the user terminal belongs; Based on the job type identifier, a list of materials matching the target job type indicated by the job type identifier is retrieved from the knowledge base; the list of materials represents the qualification information that the user to which the user terminal belongs needs to provide. The entry data set is parsed to obtain the parsing result, and the parsing result is compared with the data list to obtain the comparison result; the parsing result represents the entry data list corresponding to the entry data set, and the comparison result represents whether the entry data set is complete; If the comparison result indicates that the entry data set is incomplete, a data supplement message is sent to the user terminal, and the data supplement message includes the name of the missing entry data. If the comparison result indicates that the set of entry data is complete, the corresponding work plan is retrieved from the knowledge base based on the work type identifier, and a work instruction is generated based on the work plan and sent to the user terminal.

2. The method according to claim 1, characterized in that, The work request also includes the start and end times of the work, and the set of entry documents includes qualification certificates. The process involves parsing the set of entry documents to obtain parsing results, and comparing these results with the document list to obtain comparison results, including: The qualification certificate is parsed to obtain the qualification type and the validity period of the qualification certificate; Based on the job type identifier, the target qualification type corresponding to the job type is retrieved from the knowledge base to determine whether the qualification type is consistent with the target qualification type. Determine whether the start and end times of the operation both fall within the validity period; If the qualification type is consistent with the target qualification type, and the start and end times of the operation both fall within the validity period, then a comparison result representing the completeness of the entry data set is generated. If the qualification type is inconsistent with the target qualification type, or if the start and end times of the operation do not both fall within the validity period, a comparison result indicating that the entry data set is incomplete will be generated.

3. The method according to claim 2, characterized in that, The process of parsing the qualification certificate to obtain the qualification type and the validity period of the qualification certificate includes: Extract the text content from the qualification certificate and the location information of the text content on the qualification certificate; Compare the location information with preset location information to see if they are consistent; the preset location information represents verified standard location information. If they match, the qualification type and the validity period of the qualification certificate are determined based on the text content. If there is a discrepancy, a message indicating that the qualification certificate resolution failed will be sent to the user terminal.

4. The method according to claim 1, characterized in that, The generation of work instructions based on the work plan includes: Based on a large language model, the semantic information of the task plan is identified, and the task plan is decomposed based on the semantic information to obtain multiple task steps; the semantic information includes the execution actions of the task plan and the logical relationships between the multiple execution actions; Retrieve each of the work steps from the knowledge base and determine the risk event corresponding to each of the work steps; The work instruction is generated based on each work step and the corresponding risk event.

5. The method according to claim 4, characterized in that, The job request also includes the job environment, and determining the risk event corresponding to each job step includes: Load a preset risk map, which includes operation nodes, environmental feature nodes, and risk nodes, and the operation nodes are connected to the environmental feature nodes and the risk nodes; For a target operation step, the target operation step is mapped to a target operation node in the preset risk map, and the operation environment is mapped to a target environment feature node in the preset risk map. Based on the connection relationship between the target operation node and each risk node, the activation probability of each risk node is determined. The target operation step is any one of the plurality of operation steps, the activation probability represents the probability of occurrence of the risk event corresponding to the risk node, and the connection relationship is related to the operation environment. When the target operation steps are executed, the target risk event corresponding to the target risk node with the highest activation probability is determined.

6. The method according to claim 5, characterized in that, The step of determining the activation probability of each risk node based on the connection relationship between the target operation node and each risk node includes: Using the target operation node as the root node, traverse each risk node in the preset risk map to determine the distance between the target operation node and each risk node. Based on the distance between the target operation node and each of the risk nodes, a risk coefficient for each risk node is determined, and the risk coefficient is negatively correlated with the distance. For the target operation node, the activation probability of each risk node is determined based on the risk coefficient and risk value corresponding to each risk node; the risk value of the risk node is determined based on the risk event corresponding to the risk node.

7. The method according to claim 4, characterized in that, After generating the work instruction based on each work step and the corresponding risk event, the method includes: Determine the risk level of the risk event corresponding to each of the above work steps, and perform a weighted calculation based on the risk level of the risk event corresponding to each of the above work steps to determine the target risk level of the work plan; If the target risk level is greater than the preset risk level, retrieve the security measures corresponding to the target risk level from the knowledge base; Add the aforementioned safety measures to the aforementioned work instructions.

8. An electronic device, characterized in that, It includes a processor and a memory, the processor being coupled to the memory; the memory is used to store computer instructions, which are loaded and executed by the processor to enable the computer device to perform the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes computer-executable instructions that, when executed on a computer, cause the computer to perform the method of any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when run on an electronic device, causes the electronic device to perform the method as described in any one of claims 1 to 7.

Images