A method for intelligent filing and approval of qualification certificates

By combining digital twin simulation and reinforcement learning-driven on-chain smart contracts with multimodal intelligent data collection, federated learning, and blockchain technology, the problems of insufficient verification accuracy, cross-domain data privacy risks, and reliance on fixed rules in the qualification certificate filing and approval process have been solved, achieving end-to-end closed-loop intelligence and data empowerment across the entire chain.

CN123118016APending Publication Date: 2026-11-17THREE GORGES HI TECH INFORMATION TECH CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202611263387.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The existing qualification certificate filing and approval process suffers from insufficient verification accuracy, high privacy risks in cross-domain data interaction, reliance on fixed rules for approval decisions, and difficulty in leveraging filing data to empower talent strategies, making it difficult to achieve an end-to-end intelligent closed loop.

Method used

By employing digital twin simulation and reinforcement learning-driven on-chain smart contracts, combined with multimodal intelligent data collection, federated learning, knowledge graphs, and blockchain technology, cross-domain authoritative verification of certificates, intelligent approval decisions, and secure end-to-end evidence storage are achieved.

Benefits of technology

It has achieved authoritative cross-verification of certificate verification results, dynamic scheduling of approval decisions, and secure and available data, forming a fully intelligent closed loop from certificate verification to talent data empowerment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118016A_ABST
    Figure CN123118016A_ABST
Patent Text Reader

Abstract

This invention discloses a method for intelligent filing and approval of qualification certificates. The method extracts certificate text and anti-counterfeiting features locally through multimodal acquisition and edge computing for pre-verification; it generates an authoritative verification report by organizing multiple authoritative institution sub-nodes through a federated learning verification network in a cross-domain cross-verification mode under the data not leaving the domain mode; it intelligently associates verified certificate information based on the enterprise talent qualification knowledge graph; it constructs a digital twin mapped to the physical approval process, using it as a simulation environment for reinforcement learning to generate approval link optimization strategies; it jointly maps fuzzy indicators and strategies such as job matching degree and talent strategy priority into deterministic approval conditions through an approval decision quantification function; it dynamically schedules the approval link and executes on-chain approvals according to the deterministic approval conditions using a blockchain smart contract; it stores all process data on the blockchain and integrates privacy computing for desensitization; after approval, it synchronizes employee information and updates the strategy by returning approval records.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of qualification certificate data processing technology, and in particular to a method for intelligent filing and approval of qualification certificates. Background Technology

[0002] Qualification certificates are legal credentials that prove employees' professional abilities and qualifications in activities such as production and operation, personnel management, bidding, and qualification review. They encompass various types of certificates, including vocational qualification certificates, professional title certificates, and professional skill level certificates. With the deepening of enterprise digital transformation, qualification certificate filing and approval has become a fundamental aspect of human resource information management. Enterprises need to enter information on various qualification certificates held by employees into the system, which, after approval, is used in business scenarios such as professional title evaluation, job assignment, and subsidy calculation. However, qualification certificate filing and approval involves a wide variety of certificates from diverse sources. Verification of certificate authenticity relies on official information from authoritative institutions such as certificate issuing authorities and industry associations. The approval chain spans internal enterprise systems and external authoritative data sources. This necessitates that the technical solution for filing and approval simultaneously address a series of issues, including certificate verification, secure cross-domain data interaction, intelligent approval decision-making, and full lifecycle management of filing data. The technical problems that this application aims to solve are the core issues that enterprises face in the process of qualification certificate filing and approval, such as insufficient accuracy of certificate verification, privacy risks in cross-domain data interaction, reliance on manual and fixed rules for approval decisions, and difficulty in supporting enterprise talent strategy decisions with filing data. These problems are intertwined, making it difficult to achieve an end-to-end intelligent closed loop from certificate verification to talent value empowerment in qualification certificate filing and approval.

[0003] To address the aforementioned technical issues, existing technologies typically employ the following methods: First, regarding certificate text extraction and verification, OCR (Optical Character Recognition) technology is used to extract text from certificate attachment images. The identified structured fields, such as the certificate number, certificate holder's name, and certificate validity period, are compared with authoritative databases publicly available from the certificate issuing authority to determine the authenticity and validity of the certificate content. Some solutions further incorporate anti-counterfeiting feature recognition of the certificate image, improving verification accuracy by detecting physical features such as seals, watermarks, and printing textures. Second, regarding cross-domain data interaction, data exchange between enterprise registration systems and human resources and social security departments, industry qualification disclosure platforms, and third-party certification bodies typically employs methods such as front-end data mirroring, controlled interfaces, or dedicated line synchronization. Permission verification and access control are implemented at the boundary. Some solutions introduce federated learning or privacy-preserving computation frameworks, achieving cross-institutional data sharing through model parameter collaboration or encrypted computation with data remaining within its domain, thereby reducing the risk of original data leakage. Third, regarding the approval process, existing systems generally employ a workflow engine combined with a rule engine. Approval rules such as certificate level, approval authority, and approval level are pre-configured as fixed conditions. The approval process is then judged based on these pre-defined rules, and the approval chain is relatively fixed after configuration. Fourth, regarding data storage and security, some solutions write key data from the approval process to the blockchain. The immutability of the blockchain's distributed ledger is used to preserve and track approval records, and symmetric encryption, asymmetric encryption, or secret sharing technologies are used to protect the stored data. Fifth, regarding data utilization, existing solutions primarily use the registration data for file storage and retrieval. Some systems are linked to the company's internal employee information database, writing the certificate fields back to the corresponding fields in the employee's file after approval for subsequent use in professional title, position, and other business processes.

[0004] However, the aforementioned existing technologies still have significant shortcomings in practical applications: First, in the certificate verification process, single OCR recognition and fixed anti-counterfeiting feature verification can only complete the verification at the certificate image level. The verification results lack cross-verification from authoritative data sources across domains, making it difficult to deal with certificate forgery and content tampering, resulting in insufficient verification accuracy. Chinese patent CN121033863A discloses an anti-counterfeiting verification method for electronic certificates based on the fusion of OCR and micro-texture features. However, this scheme only compares certificate image features and does not introduce cross-verification from authoritative data from human resources and social security departments and industry qualification disclosure platforms, thus failing to identify certificates forged through high-quality printing. Second, in the cross-domain data interaction stage, direct data mirroring and controlled interface docking can easily lead to the leakage of employee information and qualification data during transmission. Existing federated learning and privacy computing solutions mostly rely on dedicated hardware such as trusted execution environments and focus on model training scenarios, making it difficult to directly address the fusion of multi-institution verification results for qualification certificates. Chinese patent CN116011590A discloses a federated learning method, apparatus, and system. However, in this scheme, data collaboration relies on the hardware deployment and unified authentication of a trusted execution environment, and does not provide a multi-dimensional fusion mechanism for verification results from multiple authoritative nodes. Third, in the approval decision-making stage, the rule-engine-based approval process solidifies approval rules into configuration items. The determination of the approval link and approver depends on preset conditions, making it impossible to achieve adaptive scheduling of the approval link by combining dynamic indicators such as job matching degree and corporate talent strategy. Chinese patent CN112749949A discloses a method for predicting flowable workflow approval paths based on the easyRule rule engine. However, in this scheme, the approval path is generated by the approver rules configured by the rule engine and the automatic approval rule prediction. The approval rules are still manually preset static conditions, and it is impossible to dynamically adjust the approval link according to employee qualifications, job requirements, and talent strategy. Fourth, regarding data storage and security, while existing blockchain-based storage solutions utilize blockchain for traceability and evidence preservation, they lack a unified desensitization and hierarchical management mechanism to balance the security protection of stored data with the availability of business data. This makes it difficult to support employee information synchronization and talent analysis while protecting privacy. Chinese patent CN113364576B discloses a blockchain-based method for encrypted data storage and sharing; however, this solution focuses on data ownership confirmation and key management, failing to provide end-to-end data desensitization, encryption, and hierarchical security management for qualification certificate filing scenarios, and also failing to achieve real-time linkage between approval results and employee information. These shortcomings collectively make it difficult for existing technologies to meet the integrated requirements of enterprise digital talent management for authoritative verification, secure cross-domain operations, intelligent decision-making, end-to-end storage, and data empowerment. Therefore, a method for intelligent qualification certificate filing and approval is needed to address these issues. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for intelligent filing and approval of qualification certificates, which aims to overcome the shortcomings of existing qualification certificate filing and approval methods, such as insufficient verification accuracy, high privacy risks of cross-domain data interaction, reliance on fixed rules for approval decisions and difficulty in empowering talent strategies with filing data. It features dynamic approval of on-chain smart contracts driven by digital twin simulation reinforcement learning, cross-domain verification of data without leaving the domain, and integrated secure evidence storage across the entire chain.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for intelligent filing and approval of qualification certificates includes the following steps: Collect qualification certificate attachments and perform pre-verification locally on the collection terminal to generate pre-verification results and structured certificate information; The structured certificate information is sent to the federated learning verification network. Each sub-node cross-verifies the information based on the local authoritative database and interacts with the model parameters and verification results to generate an authoritative verification report. Based on the enterprise's talent qualification knowledge graph, the verified certificate information is intelligently linked to generate associated filing information. The associated filing information includes fuzzy approval indicators such as job matching degree and talent strategy priority. Construct a digital twin of the filing and approval process that maps to the physical approval process, and use reinforcement learning to generate an optimization strategy for the approval process using the digital twin as a simulation environment; The fuzzy approval indicators and the approval process optimization strategy are jointly mapped into deterministic approval conditions through the approval decision quantification function; Through the smart contract for filing and approval deployed on the blockchain alliance chain, the approval link and approver are dynamically scheduled according to deterministic approval conditions, the on-chain approval is executed and the approval record is generated on the chain in real time. The entire filing process data will be stored on the blockchain and a unique storage hash value will be generated. After the approval is completed, the employee information will be intelligently synchronized and archived to the employee information database. The approval records are fed back into the digital twin to update the approval process optimization strategy.

[0007] Preferably, the step of collecting qualification certificate attachments and performing pre-verification locally on the collection terminal includes: simultaneously extracting the textual structured information and anti-counterfeiting feature information from the qualification certificate attachments by integrating OCR optical character recognition and computer vision deep learning technologies through a multimodal intelligent collection terminal, wherein the anti-counterfeiting feature information includes seal texture, digital watermark, and QR code traceability information; performing pre-verification locally on the collection terminal, wherein the pre-verification includes attachment format verification, preliminary identification of anti-counterfeiting features, and information structure parsing, and only uploading the pre-verification results and the structured certificate information to the cloud, and attachments that fail the pre-verification are directly fed back and returned for modification on the collection terminal.

[0008] Furthermore, when the acquisition terminal is a mobile terminal with computing power lower than a preset level, the acquisition terminal loads a lightweight pre-verification model and completes the pre-verification locally. When the acquisition terminal is a computer terminal or enterprise smart terminal with computing power higher than a preset level, the acquisition terminal loads a complete pre-verification model and completes the pre-verification locally. The preset level is determined according to the computing power index of the acquisition terminal processor.

[0009] Preferably, sending the structured certificate information to the federated learning verification network includes: The enterprise registration system is used as the core node, and federated learning sub-nodes are established with the databases of the human resources and social security department, the industry qualification publicity platform, and the third-party certification body. The model adopts the mode of data not leaving the domain and model co-training, without sharing the original data, only exchanging model parameters and verification results. The cloud system sends the structured certificate information to the verification network. Each sub-node completes cross-verification based on its local authoritative database and generates an authoritative verification report through a multi-dimensional verification result fusion model. Certificates that fail verification generate detailed reasons for failure and are returned. The information on forged certificates is also included in the enterprise risk qualification database.

[0010] Preferably, the generation of an authoritative verification report through a multi-dimensional verification result fusion model includes: Calculate the fusion weights for each federated learning child node: ; in For the first The authority score of each sub-node. For the first Historical verification accuracy of each child node The total number of child nodes; the authority score of the institution. The historical verification accuracy rate is determined by mapping the administrative level of the institution to which the sub-node belongs to the certificate issuance authority level. The determination is based on the proportion of samples where the verification conclusions of sub-nodes are consistent with the post-verification conclusions within a preset period, and at least one sub-node has an institutional authority score greater than zero; according to the aforementioned fusion weights. The verification results fed back by each sub-node are weighted and summed to generate an authoritative verification report.

[0011] Preferably, the intelligent association based on the enterprise talent qualification knowledge graph includes: Construct a knowledge graph of enterprise talent qualifications with employees as the core entity, link qualification certificates, professional title information, job information and subsidy rules to related entities, and establish attribute relationships and logical relationships between the entities; Based on the knowledge graph, intelligent association and automatic completion are performed on the verified certificate information, and the corresponding professional title series, job requirements and subsidy calculation rules are automatically matched. When an employee chooses to update their professional title in the system, the knowledge graph intelligently recommends matching ways to obtain the professional title and job qualifications, achieving zero-manual configuration of professional title update information.

[0012] Preferably, the reinforcement learning-based strategy for generating an optimized approval process using a digital twin as the simulation environment includes: The approval process is modeled as a state transition graph, where nodes represent approval stages and edges represent approval flow paths. Based on blockchain-stored historical approval data, the probability of each state transition and the distribution of processing time are statistically analyzed to initialize the digital twin of the filing and approval process. Approval process status Defined as a vector of the current approval node identifier, the number of tasks in progress for the candidate approver, the average processing time, the certificate level and employee rank in the associated filing information, and the talent strategy priority; Actions Defined as a combination of an approver selected from the set of candidate approvers and the next approval node; Calculate instant rewards: ; in The reward is based on the timeliness of the approval process. The reward is based on the accuracy of the approval results. The reward weight is based on the degree of alignment with the talent strategy. , , To reward the weighting coefficients, the reinforcement learning policy network is trained offline in a virtual approval process using a digital twin as the simulation environment, generating an approval link optimization policy.

[0013] Preferably, the reward amount is calculated as follows: ; in The standard processing time for the approval process. The actual processing time of the approval process, and when Not less than hour The value is zero; The value is one when the approval conclusion is consistent with the standard approval conclusion, and zero when they are inconsistent. The value is the ratio of the achievement of the talent strategy indicator corresponding to the approval result to the strategic goal, and the value range is 0 to 1; all three reward components are in the range of 0 to 1, making the magnitude of each component comparable.

[0014] Preferably, when there is no approval history data in the blockchain evidence storage, the state transition probability and processing time distribution of the filing and approval digital twin are initialized according to the node order of the preset approval process template and the preset processing time distribution, and the preset values ​​are replaced with real statistical values ​​after the approval records are accumulated back.

[0015] Preferably, the step of jointly mapping fuzzy approval indicators and approval link optimization strategies into deterministic approval conditions through an approval decision quantification function includes: Calculate job matching degree: ; in For the first accreditation entity A vector of attributes For the entity required by the job posting A vector of attributes For the first The weight of each attribute, The cosine similarity function is used. This represents the total number of attribute dimensions. Calculate talent strategy priorities: ; in For the first The actual completion values ​​of the talent strategy indicators For the first The annual target values ​​for the talent strategy indicators, For the first The weight of each talent strategy indicator Total number of strategic indicators; The job matching degree The aforementioned talent strategy priorities Link priority score output by the approval link optimization strategy Standardize to the mean and standard deviation of the registered samples respectively. , and ; Calculate the deterministic approval conditions: ; in , , The weighting coefficients and The aforementioned deterministic approval conditions The hash value is uploaded to the blockchain for use by the filing and approval smart contract.

[0016] Furthermore, the attribute weights The contribution of each attribute to the job matching degree is determined by normalization, and the contribution is calculated based on the correlation coefficient between the attribute values ​​and the job recruitment results in the historical filing samples; the strategic weight The weighting coefficient is determined by mapping the priority level of this indicator in the enterprise's talent strategy plan to the 0-1 range and then normalizing it; , , The annual target value is preset according to the priority of job matching, strategy execution, and process optimization in the enterprise approval strategy, and all of them are greater than zero; Greater than zero.

[0017] Preferably, the on-chain approval process includes: The approval rules are solidified into the execution conditions of the filing and approval smart contract. The approval rules include certificate level, employee job level, job matching degree and corporate talent strategy. The smart contract dynamically and adaptively schedules the approval process and approvers based on the deterministic approval conditions. All approval operations are recorded on the blockchain in real time to generate tamper-proof approval records. The approval operations include approval, rejection, approval opinion, and operation time. When the approver fails to process the application within the specified time, the smart contract automatically triggers a tiered reminder and reassigns the approval task according to the company's rules, ensuring that the approval process is not interrupted.

[0018] Preferably, the step of dynamically scheduling the approval process and approvers based on deterministic approval conditions includes: The aforementioned deterministic approval conditions Compared with the preset approval level range, The interval in which the approval is applied corresponds to the approval level. The number of steps and the approval level of the approval chain are determined according to the approval level. From the candidate approver sequence recommended by the approval chain optimization strategy, the approver for the current step is selected based on the minimum number of tasks in transit and the highest historical approval accuracy of each candidate approver. The selected result is then uploaded to the blockchain as an approval record.

[0019] Preferably, the step of storing the entire filing process data on the blockchain includes: The multimodal verification data of the filing application, the federated learning verification report, the knowledge graph association results and the on-chain approval records will be uploaded to the blockchain for evidence storage in real time, and a unique blockchain evidence storage hash value will be generated for each filing application. By integrating privacy-preserving computing technology to de-identify and encrypt filing data, a hierarchical security management system is constructed to achieve data usability without visibility. Once the approval is completed, the system will automatically trigger intelligent synchronization of employee information, archiving the certificate information and attachments to the corresponding fields in the employee information database according to their type. When it is necessary to update the professional title, the system will automatically complete the entry of the new professional title and reverse the status of the original default professional title, thus realizing real-time linkage between the filing information and the employee information.

[0020] Preferably, the strategy for returning approval records to the digital twin to update the approval process optimization includes: The actual processing time, approval conclusion, and approval task transfer record in the on-chain approval record are fed back to the filing and approval digital twin to update the transition probability and processing time distribution of the state transition graph; The actual processing time and approval conclusion are added to the experience playback buffer as reinforcement learning experience samples. Online fine-tuning is performed in the digital twin to update the approval link optimization strategy, so that the approval link scheduling of subsequent filing applications evolves dynamically with the approval results.

[0021] Preferably, a qualification certificate intelligent filing and approval system is provided for executing the aforementioned qualification certificate intelligent filing and approval method. The system includes: A multimodal intelligent data acquisition terminal is used to collect qualification certificate attachments and perform pre-verification locally on the acquisition terminal, generating pre-verification results and structured certificate information; The federated learning verification module is used to build a federated learning verification network, perform cross-domain cross-verification on the structured certificate information, and generate an authoritative verification report. The knowledge graph association module is used to build a knowledge graph of enterprise talent qualifications, intelligently associate verified certificate information, and generate associated filing information containing fuzzy approval indicators. The digital twin simulation and reinforcement learning module is used to construct a digital twin of the filing and approval process that maps to the physical approval process. The digital twin is used as the simulation environment to generate an approval link optimization strategy using reinforcement learning. The approval decision quantification module is used to jointly map the fuzzy approval indicators and the approval link optimization strategy into deterministic approval conditions through the approval decision quantification function. The smart contract approval module is used to deploy filing and approval smart contracts on the blockchain consortium chain, dynamically schedule the approval link and approver according to the deterministic approval conditions, and execute on-chain approval. The blockchain-based evidence storage and employee information synchronization module is used to store all data from the entire filing process on the blockchain and generate a unique evidence hash value. Once the approval is completed, it triggers intelligent synchronization of employee information. The data feedback and update module is used to feed back the approval records to the digital twin to update the approval link optimization strategy.

[0022] Preferably, the multimodal intelligent acquisition terminal sends the structured certificate information to the federated learning verification module; the federated learning verification module sends the verified certificate information to the knowledge graph association module; the knowledge graph association module sends the associated filing information to the digital twin simulation and reinforcement learning module and the approval decision quantification module respectively; the digital twin simulation and reinforcement learning module sends the generated approval link optimization strategy to the approval decision quantification module; the approval decision quantification module sends the deterministic approval conditions to the smart contract approval module; the smart contract approval module sends the approval record to the blockchain evidence storage and employee information synchronization module; the blockchain evidence storage and employee information synchronization module sends the approval record to the data feedback and update module; and the data feedback and update module writes back the updated state transition parameters to the digital twin simulation and reinforcement learning module.

[0023] Preferably, a computer device includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the aforementioned intelligent filing and approval method for qualification certificates.

[0024] Preferably, a computer-readable storage medium stores computer instructions that cause a computer to execute the aforementioned intelligent filing and approval method for qualification certificates.

[0025] The beneficial effects of this invention are as follows: 1. This invention integrates OCR optical character recognition and computer vision deep learning technologies through a multimodal intelligent acquisition terminal. It simultaneously extracts the structured text information and anti-counterfeiting features such as seal texture, digital watermark, and QR code traceability from the attachments of qualification certificates. Pre-verification, including attachment format verification, preliminary identification of anti-counterfeiting features, and structured information parsing, is performed locally at the acquisition terminal. Only the verification results and structured certificate information are uploaded to the cloud, avoiding the full upload of the original certificate data. Subsequently, through a federated learning verification network, in a data-within-a-domain and model-co-trained mode, each sub-node of the human resources and social security department database, the industry qualification publicity platform database, and the third-party certification body database conducts cross-domain cross-verification of the certificate information. An authoritative verification report is generated by weighting the data according to the authority of each sub-node institution and its historical verification accuracy. This achieves dual verification of certificate text information and anti-counterfeiting features, as well as cross-verification of authoritative data across domains. Certificates that fail verification generate detailed failure reasons, are returned, and included in the enterprise risk qualification database. This solves the shortcomings of existing technologies, such as insufficient verification accuracy and inability to identify forged and tampered certificates.

[0026] 2. This invention intelligently associates verified certificate information using an enterprise talent qualification knowledge graph, automatically matching professional title series, job requirements, and subsidy calculation rules. It constructs a knowledge graph with employees as the core entity to automatically complete registration information. A digital twin of the registration and approval process is built, mapped to the physical approval process. Driven by blockchain-stored historical approval data, the approval process is modeled as a state transition diagram. Using the digital twin as a simulation environment, a reinforcement learning offline training strategy for optimizing the approval link is employed. A quantitative function for approval decisions maps fuzzy approval indicators such as job matching degree and talent strategy priority, along with the link priority score output by the strategy, into deterministic approval conditions, which are then recorded on the blockchain. The smart contract for registration and approval on the blockchain consortium chain dynamically schedules the approval link and approvers based on these deterministic approval conditions. If an approver fails to process the application within the specified time, a tiered reminder is automatically triggered, and the application is reassigned according to enterprise rules. This upgrades approval decision-making from fixed rules to data-driven intelligent scheduling, overcoming the shortcomings of existing technologies that rely on fixed rule engines and cannot dynamically schedule approval links by combining job matching degree and enterprise talent strategy.

[0027] 3. This invention utilizes a blockchain consortium blockchain to store all data throughout the entire process, including multimodal verification data of the filing application, federated learning verification reports, knowledge graph association results, and on-chain approval records, on the blockchain in real time. A unique blockchain hash value is generated for each filing application, ensuring data immutability and end-to-end traceability. Privacy computing technology is integrated to anonymize and encrypt the filing data, and a hierarchical security management system is constructed, ensuring data is usable but not visible. Upon approval, intelligent synchronization of employee information is automatically triggered, archiving certificate information and attachments to corresponding fields in the employee information database according to type. When a professional title needs to be updated, the system automatically completes the entry of the new title and reverses the status of the original default title, achieving real-time linkage between filing information and employee information. Simultaneously, on-chain approval records are fed back to the digital twin to update state transition probabilities and used as reinforcement learning experience samples to fine-tune the approval process optimization strategy online. This forms a closed-loop system encompassing certificate verification, intelligent approval, secure storage, employee information synchronization, and talent data empowerment, solving the shortcomings of existing technologies such as data silos, privacy leakage risks, and the difficulty of using filing data to empower talent strategic decision-making. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the method flow in an embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the convergence of reinforcement learning strategy training in the digital twin for filing and approval in this embodiment of the invention; Figure 4 This is a schematic diagram illustrating that the quantitative condition G for approval decision-making in this embodiment of the invention is composed of a weighted average of job matching degree, talent strategy priority, and link priority. Figure 5 This is a schematic diagram comparing the average approval time of the reinforcement learning dynamic scheduling and fixed rule approval in this embodiment of the invention under different approval loads. Detailed Implementation

[0029] Example 1: like Figure 1 As shown, a method for intelligent filing and approval of qualification certificates includes the following steps: Collect qualification certificate attachments and perform pre-verification locally on the collection terminal to generate pre-verification results and structured certificate information; The structured certificate information is sent to the federated learning verification network. Each sub-node cross-verifies the information based on the local authoritative database and interacts with the model parameters and verification results to generate an authoritative verification report. Based on the enterprise's talent qualification knowledge graph, the verified certificate information is intelligently linked to generate associated filing information. The associated filing information includes fuzzy approval indicators such as job matching degree and talent strategy priority. Construct a digital twin of the filing and approval process that maps to the physical approval process, and use reinforcement learning to generate an optimization strategy for the approval process using the digital twin as a simulation environment; The fuzzy approval indicators and the approval process optimization strategy are jointly mapped into deterministic approval conditions through the approval decision quantification function; Through the smart contract for filing and approval deployed on the blockchain alliance chain, the approval link and approver are dynamically scheduled according to deterministic approval conditions, the on-chain approval is executed and the approval record is generated on the chain in real time. The entire filing process data will be stored on the blockchain and a unique storage hash value will be generated. After the approval is completed, the employee information will be intelligently synchronized and archived to the employee information database. The approval records are fed back into the digital twin to update the approval process optimization strategy.

[0030] Preferably, the step of collecting qualification certificate attachments and performing pre-verification locally on the collection terminal includes: simultaneously extracting the textual structured information and anti-counterfeiting feature information from the qualification certificate attachments by integrating OCR optical character recognition and computer vision deep learning technologies through a multimodal intelligent collection terminal, wherein the anti-counterfeiting feature information includes seal texture, digital watermark, and QR code traceability information; performing pre-verification locally on the collection terminal, wherein the pre-verification includes attachment format verification, preliminary identification of anti-counterfeiting features, and information structure parsing, and only uploading the pre-verification results and the structured certificate information to the cloud, and attachments that fail the pre-verification are directly fed back and returned for modification on the collection terminal.

[0031] Furthermore, when the acquisition terminal is a mobile terminal with computing power lower than a preset level, the acquisition terminal loads a lightweight pre-verification model and completes the pre-verification locally. When the acquisition terminal is a computer terminal or enterprise smart terminal with computing power higher than a preset level, the acquisition terminal loads a complete pre-verification model and completes the pre-verification locally. The preset level is determined according to the computing power index of the acquisition terminal processor.

[0032] Preferably, sending the structured certificate information to the federated learning verification network includes: The enterprise registration system is used as the core node, and federated learning sub-nodes are established with the databases of the human resources and social security department, the industry qualification publicity platform, and the third-party certification body. The model adopts the mode of data not leaving the domain and model co-training, without sharing the original data, only exchanging model parameters and verification results. The cloud system sends the structured certificate information to the verification network. Each sub-node completes cross-verification based on its local authoritative database and generates an authoritative verification report through a multi-dimensional verification result fusion model. Certificates that fail verification generate detailed reasons for failure and are returned. The information on forged certificates is also included in the enterprise risk qualification database.

[0033] Preferably, the generation of an authoritative verification report through a multi-dimensional verification result fusion model includes: Calculate the fusion weights for each federated learning child node: ; in For the first The authority score of each sub-node. For the first Historical verification accuracy of each child node The total number of child nodes; the authority score of the institution. The historical verification accuracy rate is determined by mapping the administrative level of the institution to which the sub-node belongs to the certificate issuance authority level. The determination is based on the proportion of samples where the verification conclusions of sub-nodes are consistent with the post-verification conclusions within a preset period, and at least one sub-node has an institutional authority score greater than zero; according to the aforementioned fusion weights. The verification results fed back by each sub-node are weighted and summed to generate an authoritative verification report.

[0034] Preferably, the intelligent association based on the enterprise talent qualification knowledge graph includes: Construct a knowledge graph of enterprise talent qualifications with employees as the core entity, link qualification certificates, professional title information, job information and subsidy rules to related entities, and establish attribute relationships and logical relationships between the entities; Based on the knowledge graph, intelligent association and automatic completion are performed on the verified certificate information, and the corresponding professional title series, job requirements and subsidy calculation rules are automatically matched. When an employee chooses to update their professional title in the system, the knowledge graph intelligently recommends matching ways to obtain the professional title and job qualifications, achieving zero-manual configuration of professional title update information.

[0035] Preferably, the reinforcement learning-based strategy for generating an optimized approval process using a digital twin as the simulation environment includes: The approval process is modeled as a state transition graph, where nodes represent approval stages and edges represent approval flow paths. Based on blockchain-stored historical approval data, the probability of each state transition and the distribution of processing time are statistically analyzed to initialize the digital twin of the filing and approval process. Approval process status Defined as a vector of the current approval node identifier, the number of tasks in progress for the candidate approver, the average processing time, the certificate level and employee rank in the associated filing information, and the talent strategy priority; Actions Defined as a combination of an approver selected from the set of candidate approvers and the next approval node; Calculate instant rewards: ; in The reward is based on the timeliness of the approval process. The reward is based on the accuracy of the approval results. The reward weight is based on the degree of alignment with the talent strategy. , , To reward the weighting coefficients, the reinforcement learning policy network is trained offline in a virtual approval process using a digital twin as the simulation environment, generating an approval link optimization policy.

[0036] Preferably, the reward amount is calculated as follows: ; in The standard processing time for the approval process. The actual processing time of the approval process, and when Not less than hour The value is zero; The value is one when the approval conclusion is consistent with the standard approval conclusion, and zero when they are inconsistent. The value is the ratio of the achievement of the talent strategy indicator corresponding to the approval result to the strategic goal, and the value range is 0 to 1; all three reward components are in the range of 0 to 1, making the magnitude of each component comparable.

[0037] Preferably, when there is no approval history data in the blockchain evidence storage, the state transition probability and processing time distribution of the filing and approval digital twin are initialized according to the node order of the preset approval process template and the preset processing time distribution, and the preset values ​​are replaced with real statistical values ​​after the approval records are accumulated back.

[0038] Preferably, the step of jointly mapping fuzzy approval indicators and approval link optimization strategies into deterministic approval conditions through an approval decision quantification function includes: Calculate job matching degree: ; in For the first accreditation entity A vector of attributes For the entity required by the job posting A vector of attributes For the first The weight of each attribute, The cosine similarity function is used. This represents the total number of attribute dimensions. Calculate talent strategy priorities: ; in For the first The actual completion values ​​of the talent strategy indicators For the first The annual target values ​​for the talent strategy indicators, For the first The weight of each talent strategy indicator Total number of strategic indicators; The job matching degree The aforementioned talent strategy priorities Link priority score output by the approval link optimization strategy Standardize to the mean and standard deviation of the registered samples respectively. , and ; Calculate the deterministic approval conditions: ; in , , The weighting coefficients and The aforementioned deterministic approval conditions The hash value is uploaded to the blockchain for use by the filing and approval smart contract.

[0039] Furthermore, the attribute weights The contribution of each attribute to the job matching degree is determined by normalization, and the contribution is calculated based on the correlation coefficient between the attribute values ​​and the job recruitment results in the historical filing samples; the strategic weight The weighting coefficient is determined by mapping the priority level of this indicator in the enterprise's talent strategy plan to the 0-1 range and then normalizing it; , , The annual target value is preset according to the priority of job matching, strategy execution, and process optimization in the enterprise approval strategy, and all of them are greater than zero; Greater than zero.

[0040] Preferably, the on-chain approval process includes: The approval rules are solidified into the execution conditions of the filing and approval smart contract. The approval rules include certificate level, employee job level, job matching degree and corporate talent strategy. The smart contract dynamically and adaptively schedules the approval process and approvers based on the deterministic approval conditions. All approval operations are recorded on the blockchain in real time to generate tamper-proof approval records. The approval operations include approval, rejection, approval opinion, and operation time. When the approver fails to process the application within the specified time, the smart contract automatically triggers a tiered reminder and reassigns the approval task according to the company's rules, ensuring that the approval process is not interrupted.

[0041] Preferably, the step of dynamically scheduling the approval process and approvers based on deterministic approval conditions includes: The aforementioned deterministic approval conditions Compared with the preset approval level range, The interval in which the approval is applied corresponds to the approval level. The number of steps and the approval level of the approval chain are determined according to the approval level. From the candidate approver sequence recommended by the approval chain optimization strategy, the approver for the current step is selected based on the minimum number of tasks in transit and the highest historical approval accuracy of each candidate approver. The selected result is then uploaded to the blockchain as an approval record.

[0042] Preferably, the step of storing the entire filing process data on the blockchain includes: The multimodal verification data of the filing application, the federated learning verification report, the knowledge graph association results and the on-chain approval records will be uploaded to the blockchain for evidence storage in real time, and a unique blockchain evidence storage hash value will be generated for each filing application. By integrating privacy-preserving computing technology to de-identify and encrypt filing data, a hierarchical security management system is constructed to achieve data usability without visibility. Once the approval is completed, the system will automatically trigger intelligent synchronization of employee information, archiving the certificate information and attachments to the corresponding fields in the employee information database according to their type. When it is necessary to update the professional title, the system will automatically complete the entry of the new professional title and reverse the status of the original default professional title, thus realizing real-time linkage between the filing information and the employee information.

[0043] Preferably, the strategy for returning approval records to the digital twin to update the approval process optimization includes: The actual processing time, approval conclusion, and approval task transfer record in the on-chain approval record are fed back to the filing and approval digital twin to update the transition probability and processing time distribution of the state transition graph; The actual processing time and approval conclusion are added to the experience playback buffer as reinforcement learning experience samples. Online fine-tuning is performed in the digital twin to update the approval link optimization strategy, so that the approval link scheduling of subsequent filing applications evolves dynamically with the approval results.

[0044] like Figure 2 As shown, a qualification certificate intelligent filing and approval system is used to execute the aforementioned qualification certificate intelligent filing and approval method. The system includes: A multimodal intelligent data acquisition terminal is used to collect qualification certificate attachments and perform pre-verification locally on the acquisition terminal, generating pre-verification results and structured certificate information; The federated learning verification module is used to build a federated learning verification network, perform cross-domain cross-verification on the structured certificate information, and generate an authoritative verification report. The knowledge graph association module is used to build a knowledge graph of enterprise talent qualifications, intelligently associate verified certificate information, and generate associated filing information containing fuzzy approval indicators. The digital twin simulation and reinforcement learning module is used to construct a digital twin of the filing and approval process that maps to the physical approval process. The digital twin is used as the simulation environment to generate an approval link optimization strategy using reinforcement learning. The approval decision quantification module is used to jointly map the fuzzy approval indicators and the approval link optimization strategy into deterministic approval conditions through the approval decision quantification function. The smart contract approval module is used to deploy filing and approval smart contracts on the blockchain consortium chain, dynamically schedule the approval link and approver according to the deterministic approval conditions, and execute on-chain approval. The blockchain-based evidence storage and employee information synchronization module is used to store all data from the entire filing process on the blockchain and generate a unique evidence hash value. Once the approval is completed, it triggers intelligent synchronization of employee information. The data feedback and update module is used to feed back the approval records to the digital twin to update the approval link optimization strategy.

[0045] Preferably, the multimodal intelligent acquisition terminal sends the structured certificate information to the federated learning verification module; the federated learning verification module sends the verified certificate information to the knowledge graph association module; the knowledge graph association module sends the associated filing information to the digital twin simulation and reinforcement learning module and the approval decision quantification module respectively; the digital twin simulation and reinforcement learning module sends the generated approval link optimization strategy to the approval decision quantification module; the approval decision quantification module sends the deterministic approval conditions to the smart contract approval module; the smart contract approval module sends the approval record to the blockchain evidence storage and employee information synchronization module; the blockchain evidence storage and employee information synchronization module sends the approval record to the data feedback and update module; and the data feedback and update module writes back the updated state transition parameters to the digital twin simulation and reinforcement learning module.

[0046] Preferably, a computer device includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the aforementioned intelligent filing and approval method for qualification certificates.

[0047] Preferably, a computer-readable storage medium stores computer instructions that cause a computer to execute the aforementioned intelligent filing and approval method for qualification certificates.

[0048] Example 2: This embodiment uses a large group enterprise as the application scenario. The group has over 3,000 employees in various positions including engineering, R&D, and management. Each year, the group needs to register and approve the professional qualification certificates and professional title certificates held by its employees for job appointments, title promotions, and subsidy calculations. The group's internal registration system needs to conduct cross-domain verification and data collaboration with the local human resources and social security department, industry qualification disclosure platform, and third-party certification bodies. The experimental platform uses MATLAB R2023a and deploys six functional modules: a multimodal intelligent acquisition terminal simulation module, a federated learning verification network simulation module, an enterprise talent qualification knowledge graph association module, a registration and approval digital twin and reinforcement learning strategy network, an approval decision quantification module, and a blockchain smart contract approval module. The parameter settings are as follows: the institutional authority of the four types of authoritative institution sub-nodes. The historical verification accuracy rates are as follows: Enterprise Node 1.0, Human Resources and Social Security Department 0.9, Industry Qualification Publicity Platform 0.85, and Third-Party Certification Agency 0.75. The values ​​are 0.96, 0.95, 0.93, and 0.90 respectively, with the fusion weights calculated as follows: Calculations yield the following weightings: enterprise nodes 0.293, human resources and social security departments 0.261, industry qualification disclosure platforms 0.241, and third-party certification bodies 0.205. (This is the weighting coefficient for reinforcement learning rewards.) Take 0.4, Take 0.3, Take 0.3; Quantitative weighting coefficient for approval decisions Take 0.4, Take 0.35, Take 0.25; Deterministic approval conditions A value of not less than 0.7 corresponds to a first-level approval process that includes two steps and a standard processing time of 8 hours. A value not less than 0.4 and less than 0.7 corresponds to a level-two approval process with three steps and a standard processing time of 12 hours. A value less than 0.4 corresponds to a three-tier approval process with four steps and a standard processing time of 16 hours. The strategy network learning rate is 0.001, the batch size is 64, the exploration rate linearly decreases from 1.0 to 0.1, the experience replay buffer capacity is 2000 entries, and the digital twin undergoes 5000 offline training rounds. The total simulation sample size is 400 filing applications, divided into 5 batches of 80 each. The verification confidence level is subject to a Gaussian perturbation with a mean of 0 and a standard deviation of 0.03. The processing time follows a normal distribution with a mean of 10% of the standard processing time and a standard deviation of 10%.

[0049] The comparison method is the existing single-modal OCR verification combined with fixed rule approval method, which only performs OCR recognition on the certificate text and compares it with a single human resources and social security database. The approval is assigned to a fixed approver according to a preset node sequence, without performing anti-counterfeiting feature verification, cross-domain multi-source fusion, dynamic approval scheduling and strategy optimization.

[0050] The implementation process is as follows: First, 400 filing application samples covering the verification results of four types of institutions are generated, and a reinforcement learning strategy is trained offline in a digital twin. Then, the strategy and approval decision quantification function are applied to simulate the entire process of data collection, verification, association, quantification, approval, evidence storage, employee synchronization, and strategy feedback for five batches. The accuracy rate of fusion verification, average approval time, overdue transfer rate, and accuracy rate of professional title synchronization for each batch are statistically analyzed. Finally, a comparison method is run under the same conditions, and the comparison data are statistically analyzed. The data are shown in Tables 1 and 2 below. Table 1: Batch operation data of this method;

[0051] Table 2: Comparison data between our method and the comparative method under different loads;

[0052] like Figure 3 As shown, this plan will provide immediate rewards. All three components are normalized to the 0-1 range, making the reward magnitudes for approval timeliness, approval accuracy, and talent strategy alignment comparable. The policy network is trained offline in a digital twin virtual approval environment, and the average cumulative reward steadily increases with the number of training rounds and converges to a stable level. In contrast, the proposed solution directly accumulates the unnormalized reward components during training, resulting in the approval timeliness component being much larger than the others. This leads to an overemphasis on shortening the timeliness while neglecting approval quality and strategic alignment, resulting in slow convergence and significant fluctuations. This figure illustrates that the proposed solution, through the combination of reward component normalization and a digital twin simulation environment, obtains a fast-converging and highly stable approval chain optimization strategy without the need for trial and error in a real approval system, thus overcoming the technical bottleneck of reinforcement learning's inability to conduct online trial and error in real approval scenarios.

[0053] like Figure 4 As shown, the approval decision quantification function will determine the job matching degree. Talent strategy priority Link priority score output by the approval link optimization strategy Joint mapping as deterministic approval conditions The process is shown in the graph. The stacked bars represent the weighted contributions of the three components in each filing application sample, and the black line represents the deterministic approval conditions. The two dashed lines represent the approval level thresholds of 0.4 and 0.7. Existing smart contract automatic approval schemes can only handle deterministic fixed conditions such as certificate levels, and cannot evaluate fuzzy indicators such as job matching degree and talent strategy. This scheme uses a quantization function to uniformly map fuzzy indicators and reinforcement learning strategy outputs into deterministic values, enabling smart contracts to evaluate based on... The range of approval levels determines the number of approval steps and the approval level, enabling on-chain verifiable dynamic approval scheduling and solving the technical problem that smart contracts cannot execute fuzzy approval conditions.

[0054] like Figure 5 As shown, this figure compares the average approval time of this solution with that of the existing fixed-rule approval under different batch loads of approval tasks. Fixed-rule approval assigns approvers according to a preset node sequence. Under high load, some approvers have tasks in transit, causing the average approval time to rise rapidly with the load. This solution, trained in a digital twin simulation environment, dynamically assigns approvers based on the number of tasks in transit for candidate approvers and their historical approval accuracy. When the load increases, it automatically distributes tasks evenly to approvers with low loads, resulting in an average approval time that increases approximately linearly and smoothly with the load. This figure illustrates that this solution, through dynamic scheduling of the approval chain driven by a reinforcement learning strategy, maintains the stability of approval timeliness under fluctuating approval load scenarios, avoids process blockage caused by approver task overload, and improves the throughput capacity of the approval system under high load conditions.

[0055] As shown in Tables 1 and 2 above, the superiority of this method compared to existing technologies lies in three dimensions: verification, scheduling, and closed-loop processing. In terms of verification, this method extracts anti-counterfeiting features such as seal texture, digital watermark, and QR code traceability through multimodal intelligent data acquisition. It organizes four types of authoritative institutions to conduct federated learning cross-domain cross-verification under a data-within-the-domain mode, and generates an authoritative verification report based on the fusion weight determined by the institution's authority and historical accuracy. Table 2 shows that this method still achieves a verification accuracy of 97.8% under a high load of 200 applications, while existing single-modal OCR verification methods, lacking anti-counterfeiting feature verification and cross-domain authoritative cross-verification, have an accuracy of only 89.1%, which continues to decline with increasing load. This indicates that this method effectively solves the problem of missed detection of forged and tampered certificates. In terms of approval scheduling, this method uses a digital twin as a simulation environment to train reinforcement learning strategies offline. Through the approval decision quantification function, fuzzy indicators such as job matching degree and talent strategy priority are jointly mapped with the strategy output to deterministic approval conditions to drive the dynamic scheduling of smart contracts. Table 2 shows that under a load of 200 applications, the average approval time of this method is 13.5 hours and the overdue transfer rate is 2.6%, while the existing fixed rule approval method has an average approval time as high as 31.5 hours and an overdue transfer rate of 21.0%. This indicates that this method avoids the overload of approvers through approval load balancing and dynamic transfer, and maintains stable approval timeliness under high load scenarios. In terms of system capabilities, this method forms a closed loop across the entire chain, encompassing verification, approval, evidence storage, employee synchronization, and strategy feedback. Table 1 shows that the accuracy rate of professional title synchronization remained between 98.6% and 99.4% across five batches, while the overdue transfer rate was controlled between 1.6% and 2.2%. This indicates that while ensuring data security and full-chain traceability, this method also achieves stable operation of the approval process and continuous empowerment of filing data, meeting the integrated requirements of enterprise digital talent management for authoritative verification, secure cross-domain operations, intelligent decision-making, full-chain evidence storage, and data empowerment.

Claims

1. A method for intelligent recordation and examination of qualification certificates, characterized in that, Includes the following steps: Collect qualification certificate attachments and perform pre-verification locally on the collection terminal to generate pre-verification results and structured certificate information; The structured certificate information is sent to the federated learning verification network. Each sub-node cross-verifies the information based on the local authoritative database and interacts with the model parameters and verification results to generate an authoritative verification report. Based on the enterprise's talent qualification knowledge graph, the verified certificate information is intelligently linked to generate associated filing information. The associated filing information includes fuzzy approval indicators such as job matching degree and talent strategy priority. Construct a digital twin of the filing and approval process that maps to the physical approval process, and use reinforcement learning to generate an optimization strategy for the approval process using the digital twin as a simulation environment; The fuzzy approval indicators and the approval process optimization strategy are jointly mapped into deterministic approval conditions through the approval decision quantification function; Through the smart contract for filing and approval deployed on the blockchain alliance chain, the approval link and approver are dynamically scheduled according to deterministic approval conditions, the on-chain approval is executed and the approval record is generated on the chain in real time. The entire filing process data will be stored on the blockchain and a unique storage hash value will be generated. After the approval is completed, the employee information will be intelligently synchronized and archived to the employee information database. The approval records are fed back into the digital twin to update the approval process optimization strategy.

2. The method for intelligent recordation and examination and approval of a qualification certificate according to claim 1, characterized in that, The process of collecting qualification certificate attachments and performing pre-verification locally on the collection terminal includes: simultaneously extracting the textual structured information and anti-counterfeiting feature information from the qualification certificate attachments by integrating OCR optical character recognition and computer vision deep learning technologies through a multimodal intelligent collection terminal. The anti-counterfeiting feature information includes seal texture, digital watermark, and QR code traceability information. Pre-verification is performed locally on the collection terminal, which includes attachment format verification, preliminary identification of anti-counterfeiting features, and information structure parsing. Only the pre-verification results and the structured certificate information are uploaded to the cloud. Attachments that fail the pre-verification are directly fed back and returned for modification on the collection terminal.

3. The method for intelligent recordation and examination and approval of a qualification certificate according to claim 1, characterized in that, Sending the structured certificate information to the federated learning verification network includes: The enterprise registration system is used as the core node, and federated learning sub-nodes are established with the databases of the human resources and social security department, the industry qualification publicity platform, and the third-party certification body. The model adopts the mode of data not leaving the domain and model co-training, without sharing the original data, only exchanging model parameters and verification results. The cloud system sends the structured certificate information to the verification network. Each sub-node completes cross-verification based on its local authoritative database and generates an authoritative verification report through a multi-dimensional verification result fusion model. Certificates that fail verification generate detailed reasons for failure and are returned. The information on forged certificates is also included in the enterprise risk qualification database.

4. The method for intelligent recordation and examination and approval of a qualification certificate according to claim 3, characterized in that, The generation of an authoritative verification report through a multi-dimensional verification result fusion model includes: Calculate the fusion weights for each federated learning child node: ; in For the first The authority score of each sub-node. For the first Historical verification accuracy of each child node The total number of child nodes; the authority score of the institution. The historical verification accuracy rate is determined by mapping the administrative level of the institution to which the sub-node belongs to the certificate issuance authority level. The determination is based on the proportion of samples where the verification conclusions of sub-nodes are consistent with the post-verification conclusions within a preset period, and at least one sub-node has an institutional authority score greater than zero; according to the aforementioned fusion weights. The verification results fed back by each sub-node are weighted and summed to generate an authoritative verification report.

5. The method for intelligent filing and approval of qualification certificates according to claim 1, characterized in that, The intelligent association based on the enterprise talent qualification knowledge graph includes: Construct a knowledge graph of enterprise talent qualifications with employees as the core entity, link qualification certificates, professional title information, job information and subsidy rules to related entities, and establish attribute relationships and logical relationships between the entities; Based on the knowledge graph, intelligent association and automatic completion are performed on the verified certificate information, and the corresponding professional title series, job requirements and subsidy calculation rules are automatically matched. When an employee chooses to update their professional title in the system, the knowledge graph intelligently recommends matching ways to obtain the professional title and job qualifications, achieving zero-manual configuration of professional title update information.

6. The method for intelligent filing and approval of qualification certificates according to claim 1, characterized in that, The reinforcement learning-based optimization strategy for generating approval processes using a digital twin as the simulation environment includes: The approval process is modeled as a state transition graph, where nodes represent approval stages and edges represent approval flow paths. Based on blockchain-stored historical approval data, the probability of each state transition and the distribution of processing time are statistically analyzed to initialize the digital twin of the filing and approval process. Approval process status Defined as a vector of the current approval node identifier, the number of tasks in progress for the candidate approver, the average processing time, the certificate level and employee rank in the associated filing information, and the talent strategy priority; Actions Defined as a combination of an approver selected from the set of candidate approvers and the next approval node; Calculate instant rewards: ; in The reward is based on the timeliness of the approval process. The reward is based on the accuracy of the approval results. The reward weight is based on the degree of alignment with the talent strategy. , , To reward the weighting coefficients, the reinforcement learning policy network is trained offline in a virtual approval process using a digital twin as the simulation environment, generating an approval link optimization policy.

7. The method for intelligent filing and approval of qualification certificates according to claim 6, characterized in that, The reward amount is calculated as follows: ; in The standard processing time for the approval process. The actual processing time of the approval process, and when Not less than hour The value is zero; The value is one when the approval conclusion is consistent with the standard approval conclusion, and zero when they are inconsistent. The value is the ratio of the achievement rate of the talent strategy indicator corresponding to the approval result to the strategic goal, and the value range is from 0 to 1; All three reward components are in the range of 0 to 1, making the magnitudes of each component comparable.

8. The method for intelligent filing and approval of qualification certificates according to claim 6, characterized in that, When there is no approval history data in the blockchain evidence storage, the state transition probability and processing time distribution of the filing and approval digital twin are initialized according to the node order of the preset approval process template and the preset processing time distribution, and the preset values ​​are replaced with real statistical values ​​after the approval records are accumulated back.

9. The method for intelligent filing and approval of qualification certificates according to claim 1, characterized in that, The step of jointly mapping fuzzy approval indicators and approval link optimization strategies into deterministic approval conditions through an approval decision quantification function includes: Calculate job matching degree: ; in For the first accreditation entity A vector of attributes For the entity required by the job posting A vector of attributes For the first The weight of each attribute, The cosine similarity function is used. This represents the total number of attribute dimensions. Calculate talent strategy priorities: ; in For the first The actual completion values ​​of the talent strategy indicators For the first The annual target values ​​for the talent strategy indicators, For the first The weight of each talent strategy indicator Total number of strategic indicators; The job matching degree The aforementioned talent strategy priorities Link priority score output by the approval link optimization strategy Standardize to the mean and standard deviation of the registered samples respectively. , and ; Calculate the deterministic approval conditions: ; in , , The weighting coefficients and The aforementioned deterministic approval conditions The hash value is uploaded to the blockchain for use by the filing and approval smart contract.

10. The method for intelligent filing and approval of qualification certificates according to claim 1, characterized in that, The on-chain approval process includes: The approval rules are solidified into the execution conditions of the filing and approval smart contract. The approval rules include certificate level, employee job level, job matching degree and corporate talent strategy. The smart contract dynamically and adaptively schedules the approval process and approvers based on the deterministic approval conditions. All approval operations are recorded on the blockchain in real time to generate tamper-proof approval records. The approval operations include approval, rejection, approval opinion, and operation time. When the approver fails to process the application within the specified time, the smart contract automatically triggers a tiered reminder and reassigns the approval task according to the company's rules, ensuring that the approval process is not interrupted.

11. The method for intelligent filing and approval of qualification certificates according to claim 10, characterized in that, The dynamic scheduling of the approval process and approvers based on deterministic approval conditions includes: The aforementioned deterministic approval conditions Compared with the preset approval level range, The interval in which the approval is applied corresponds to the approval level. The number of steps and the approval level of the approval chain are determined according to the approval level. From the candidate approver sequence recommended by the approval chain optimization strategy, the approver for the current step is selected based on the minimum number of tasks in transit and the highest historical approval accuracy of each candidate approver. The selected result is then uploaded to the blockchain as an approval record.

12. The method for intelligent filing and approval of qualification certificates according to claim 1, characterized in that, The process of storing all registration process data on the blockchain includes: The multimodal verification data of the filing application, the federated learning verification report, the knowledge graph association results and the on-chain approval records will be uploaded to the blockchain for evidence storage in real time, and a unique blockchain evidence storage hash value will be generated for each filing application. By integrating privacy-preserving computing technology to de-identify and encrypt filing data, a hierarchical security management system is constructed to achieve data usability without visibility. Once the approval is completed, the system will automatically trigger intelligent synchronization of employee information, archiving the certificate information and attachments to the corresponding fields in the employee information database according to their type. When it is necessary to update the professional title, the system will automatically complete the entry of the new professional title and reverse the status of the original default professional title, thus realizing real-time linkage between the filing information and the employee information.

13. The method for intelligent filing and approval of qualification certificates according to claim 1, characterized in that, The optimization strategy for returning approval records to the digital twin to update the approval process includes: The actual processing time, approval conclusion, and approval task transfer record in the on-chain approval record are fed back to the filing and approval digital twin to update the transition probability and processing time distribution of the state transition graph; The actual processing time and approval conclusion are added to the experience playback buffer as reinforcement learning experience samples. Online fine-tuning is performed in the digital twin to update the approval link optimization strategy, so that the approval link scheduling of subsequent filing applications evolves dynamically with the approval results.

14. A qualification certificate intelligent filing and approval system, used to execute the qualification certificate intelligent filing and approval method according to any one of claims 1-13, characterized in that, The system includes: A multimodal intelligent data acquisition terminal is used to collect qualification certificate attachments and perform pre-verification locally on the acquisition terminal, generating pre-verification results and structured certificate information; The federated learning verification module is used to build a federated learning verification network, perform cross-domain cross-verification on the structured certificate information, and generate an authoritative verification report. The knowledge graph association module is used to build a knowledge graph of enterprise talent qualifications, intelligently associate verified certificate information, and generate associated filing information containing fuzzy approval indicators. The digital twin simulation and reinforcement learning module is used to construct a digital twin of the filing and approval process that maps to the physical approval process. The digital twin is used as the simulation environment to generate an approval link optimization strategy using reinforcement learning. The approval decision quantification module is used to jointly map the fuzzy approval indicators and the approval link optimization strategy into deterministic approval conditions through the approval decision quantification function. The smart contract approval module is used to deploy filing and approval smart contracts on the blockchain consortium chain, dynamically schedule the approval link and approver according to the deterministic approval conditions, and execute on-chain approval. The blockchain-based evidence storage and employee information synchronization module is used to store all data from the entire filing process on the blockchain and generate a unique evidence hash value. Once the approval is completed, it triggers intelligent synchronization of employee information. The data feedback and update module is used to feed back the approval records to the digital twin to update the approval link optimization strategy.

15. The intelligent filing and approval system for qualification certificates according to claim 14, characterized in that, The multimodal intelligent acquisition terminal sends the structured certificate information to the federated learning verification module. The federated learning verification module sends the verified certificate information to the knowledge graph association module. The knowledge graph association module sends the associated filing information to the digital twin simulation and reinforcement learning module and the approval decision quantification module, respectively. The digital twin simulation and reinforcement learning module sends the generated approval link optimization strategy to the approval decision quantification module. The approval decision quantification module sends the deterministic approval conditions to the smart contract approval module. The smart contract approval module sends the approval record to the blockchain evidence storage and employee information synchronization module. The blockchain evidence storage and employee information synchronization module sends the approval record to the data feedback and update module. The data feedback and update module writes the updated state transition parameters back to the digital twin simulation and reinforcement learning module.

16. A computer device, characterized in that, It includes a memory and a processor, which are interconnected and communicate with each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the intelligent filing and approval method for qualification certificates as described in any one of claims 1 to 13.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute any one of claims 1 to 13 of the intelligent filing and approval method for qualification certificates.

Citation Information

Patent Citations

  • Method for predicting floable workflow approval path based on eyeRule rule engine configuration

    CN112749949A

  • A blockchain-based method for data encryption, storage, and sharing.

    CN113364576B

  • Federal learning method, device and system

    CN116011590A

  • Electronic certificate anti-counterfeiting storage verification method based on optical character recognition (OCR) and micro texture feature fusion

    CN121033863A