A power grid enterprise employee capability evaluation and career development planning method and system

By constructing a time-series heterogeneous capability knowledge graph and a self-supervised Transformer model, the problems of data heterogeneity and risk adaptation in the capability assessment of power grid enterprise employees were solved, and more accurate and secure career development planning was achieved.

CN122656475BActive Publication Date: 2026-11-10JIANGSU YIQICE NETWORK TECH CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611150445.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-07-31
Publication Date
2026-11-10
Estimated Expiration
2046-07-31

AI Technical Summary

Technical Problem

The existing employee competency assessment methods of power grid companies are difficult to uniformly process heterogeneous data in multiple business systems, and cannot be adapted to job positions, tasks and risk scenarios, resulting in inaccurate assessment results and a lack of credibility and security, which affects the reliability and security of career development planning.

Method used

Employing a time-series heterogeneous capability knowledge graph based on job task-equipment risk constraints, a self-supervised heterogeneous graph Transformer, uncertainty calibration, and multi-objective programming techniques, this approach generates scenario-based capability vectors and job suitability by uniformly mapping employee capability evidence data, and generates career development paths under credible intervals and safety risk constraints.

Benefits of technology

It improves the scenario adaptability and timeliness of employee competency assessment, ensures the credibility and security of assessment results, and generates more accurate career development paths.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122656475B_ABST
    Figure CN122656475B_ABST
Patent Text Reader

Abstract

The application discloses a power grid enterprise employee capacity evaluation and career development planning method and system, belongs to the field of power grid enterprise employee capacity data processing and career development planning, and aims at solving the problems of scattered capacity evidence data, inconsistent caliber, difficult unified representation of capacity labels, and difficult dynamic evaluation of post adaptation, constructing a time sequence heterogeneous capacity knowledge graph, extracting a conditional subgraph, generating a capacity vector by using a heterogeneous graph Transformer with a dynamic evidence weight bias, combining a credible interval and a multi-objective planning to output a career development path, and realizing the scene, dynamic and risk constraint of capacity evaluation, short board identification, post adaptation and development planning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of data processing and career development planning for the competence of employees in power grid companies, and particularly to a method and system for assessing the competence and career development planning of employees in power grid companies. Background Technology

[0002] With the advancement of digital management and production operation systems in power grid enterprises, business systems such as training, examinations, qualifications, work records, defect handling, and safety control are continuously accumulating data related to employee capabilities. This data reflects employees' knowledge, skills, experience, and compliance status under different job tasks, equipment types, and risk scenarios, serving as an important basis for conducting job competency assessments, training arrangements, and career development planning.

[0003] Existing employee competency assessment methods typically rely on manual scoring, fixed indicator tables, or statistical results from a single system, making it difficult to simultaneously handle heterogeneous data from multiple business systems, time-varying data, and differences in risk scenarios. On the one hand, inconsistencies in data definitions, update cycles, and competency label definitions across different systems make it difficult to quantify the same competency item uniformly. On the other hand, fixed indicators are difficult to combine with job positions, tasks, equipment, and risk levels for conditional assessment, easily overlooking employees' competency gaps and job mismatches in specific work scenarios.

[0004] Furthermore, existing career development plans typically translate assessment results directly into training recommendations, lacking unified constraints on the credibility of assessment results, sufficiency of evidence, training resources, certification period, and job safety risks. When evidence is insufficient, the risk level is high, or the assessment results are unstable, it is difficult to promptly differentiate between different approaches such as supplementing evidence, arranging training, and temporarily not recommending the program, affecting the reliability and security of employee development path generation.

[0005] Therefore, there is a need for a method and system for assessing the competence and career development planning of employees in power grid companies that can address the shortcomings of existing technologies. Summary of the Invention

[0006] One objective of this invention is to propose a method and system for assessing the competence and career development planning of employees in power grid enterprises. Addressing the problems of scattered employee competence data, inconsistent data definitions, difficulty in unifying competence labels, and difficulty in adapting assessment results to job tasks and risk scenarios in existing technologies, this invention proposes a technical solution based on a time-series heterogeneous competence knowledge graph constrained by job tasks and equipment risks, a self-supervised heterogeneous graph Transformer, uncertainty calibration, and multi-objective career planning. This enables employee competence evidence to be temporally correlated and scenario-based represented within a unified graph, and generates competence scores, competence gaps, job suitability, and career development paths under credible intervals and safety risk constraints. This invention effectively improves the scenario adaptability, timeliness, and safety reliability of competence assessment.

[0007] This invention provides a method for assessing the competence and career development planning of employees in power grid enterprises, comprising: S1, collecting employee competence evidence data from training, examination, qualification, work records, defect handling, and safety management systems; uniformly mapping data sources, time, employees, positions, tasks, equipment, risk levels, and competence tags to generate a time-series heterogeneous competence knowledge graph containing business event nodes and certificate-holding edges; S2, generating scenario query conditions containing position, task, equipment, and risk level for the employees to be assessed and target positions; extracting conditional subgraphs and generating dynamic evidence weights for evidence edges; S3, inputting the conditional subgraphs into the trained heterogeneous graph Transformer. r, based on relational encoding, time encoding, scenario query encoding, and dynamic evidence weights, forms an attention bias and outputs a scenario-based employee capability vector and a job capability baseline vector; S4, performs dimension alignment and difference calculation on the two vectors to generate capability scores, capability deficiency sets, job suitability, and their confidence intervals; S5, inputs the capability deficiency set, job suitability, confidence interval, training resources, certification cycle, and risk level into a multi-objective programming model. When the job suitability reaches the job suitability threshold, the confidence interval width does not exceed the confidence interval threshold, and the risk level is not higher than the job safety threshold, a career development path is output; otherwise, a supplementary certification instruction and / or training instruction and / or a temporary no-recommendation instruction are output.

[0008] Optionally, S1 includes:

[0009] Generate evidence records for each piece of employee capability evidence data. The evidence records include the data source identifier, the time of generation, the time of update, the business object identifier, the event result, and the risk level.

[0010] Map the same employee, position, task, equipment, and capability tags to a unified identifier according to the preset field dictionary;

[0011] Using employees, positions, tasks, equipment, risk scenarios, capability tags, courses, certificates, and business events as nodes, and taking positions, participation, associated equipment, corresponding risks, proven capabilities, required capabilities, completed courses, and held certificates as edges, and writing timestamps according to the time of generation, a temporally heterogeneous capability knowledge graph is obtained.

[0012] Optionally, S2 includes:

[0013] The dynamic evidence weight is obtained by normalizing and weighting the data source reliability, time decay coefficient, cross-system consistency coefficient, and risk scenario relevance.

[0014] The reliability of the data source is determined by a preset data source level mapping table;

[0015] The time decay coefficient is determined based on the interval between the assessment time and the evidence generation time, as well as the validity period parameter corresponding to the position.

[0016] The cross-system consistency coefficient is determined based on the ratio of the number of consistent evidence results for the same capability label in different systems to the number of comparisons.

[0017] The relevance of risk scenarios is determined based on the number of matching items between the task, device, and risk level of the evidence event and the scenario query conditions;

[0018] Dynamic evidence weights are used as edge attributes of evidence edges in the conditional subgraph;

[0019] Furthermore, the dynamic evidence weight is updated when a new evidence record is added, specifically including: retrieving historical evidence records under the same employee, the same capability tag, and the same risk scenario based on the business object identifier of the new evidence record;

[0020] The cross-system consistency coefficient is adjusted based on the consistency between newly added evidence records and historical evidence records.

[0021] The time decay coefficient is recalculated based on the time of generation of the newly added evidence records;

[0022] The corrected cross-system consistency coefficient and the recalculated time decay coefficient are renormalized with the data source reliability and risk scenario correlation to obtain the updated dynamic evidence weight, which is then written back to the corresponding evidence edge.

[0023] Optionally, S3 includes:

[0024] After concatenating node type embedding, relation type embedding, time interval embedding, and scenario query embedding, the data is mapped to query vector, key vector, and value vector.

[0025] The dynamic evidence weights are converted into attention bias terms using a monotonic mapping table, and the attention bias terms are added to the dot product of the query vector and the key vector corresponding to the same evidence edge.

[0026] Aggregate the value vectors of neighboring nodes according to node type, and output the scenario-based employee ability vector and job ability benchmark vector;

[0027] Furthermore, the heterogeneous graph Transformer is obtained through self-supervised training, and the training samples are generated from a historical time-series heterogeneous capability knowledge graph;

[0028] The training tasks include occlusion node type recovery, occlusion relationship type recovery, future business event outcome prediction, and capability label consistency judgment.

[0029] The training loss is obtained by weighted summation of the losses of each training task and the evidence weight bias regularization term;

[0030] After training, the model parameters are fixed, and the trained model is used for the vector output in step S3.

[0031] Optionally, S4 includes:

[0032] Based on the dimensional order of the job competency benchmark vector, the scenario-based employee competency vector is dimensionally completed and normalized, and the dimensional difference vector is calculated.

[0033] Input the dimension-wise difference vector into the scoring mapping table to obtain the ability score and set of ability shortcomings for each ability tag;

[0034] Input the dimension-wise difference vector into the fit degree mapping model to obtain the job fit degree;

[0035] The non-compliance score is constructed by mapping the observed capability indicators obtained from subsequent work closed-loop quality, safety violation rate, defect recurrence rate or test retest scores with the prediction residuals of the prediction results. The quantile values ​​of the non-compliance score distribution are taken under the target coverage, and the confidence intervals of capability score and job suitability are generated from the quantile values.

[0036] Optionally, S5 includes:

[0037] Establish reachability relationships between the capability tags in the set of capability gaps and courses, certificates, job tasks and certification cycles to generate candidate development actions;

[0038] For each candidate development action, calculate the required capacity supplementation, training duration, certification time, and risk constraint satisfaction.

[0039] When the job fit reaches the job fit threshold, the confidence interval width does not exceed the confidence interval threshold, and the risk level is not higher than the job safety threshold, a sequence of candidate development actions is selected according to the weighted targets of capability supplementation, training duration, and certification time to obtain a career development path.

[0040] Furthermore, the supplementary certification instruction, training instruction, or temporary non-recommendation instruction are generated according to objective conditions and the priority of the main instruction: when the risk level is higher than the job safety threshold or the job suitability is lower than the job suitability threshold, the temporary non-recommendation condition is met; when the width of the confidence interval exceeds the confidence interval threshold and the number of missing evidence reaches the supplementary certification quantity threshold, the supplementary certification condition is met; when the width of the confidence interval does not exceed the confidence interval threshold and the set of capability shortcomings is not empty, the training condition is met.

[0041] If the conditions for temporary non-recommendation are met, a temporary non-recommendation instruction is generated as the main instruction, and the unmet risk level, job task, and competency label are output; if the conditions for supplementary certification and / or training are met at the same time, corresponding supplementary certification instructions and / or training instructions are generated as subsidiary instructions and written into the same instruction queue.

[0042] If the conditions for not recommending certification are not met but the conditions for supplementary certification are met, a supplementary certification instruction will be generated as the main instruction, and the data source identifier and capability tag to be supplemented will be output. If the conditions for not recommending certification and supplementary certification are not met but the conditions for training are met, a training instruction will be generated as the main instruction, and the course identifier, certificate identifier and completion deadline will be output.

[0043] On the other hand, the present invention also provides a system for assessing the competence and career development planning of employees in power grid enterprises, comprising:

[0044] The graph construction module is used to collect employee competency evidence data and generate a time-series heterogeneous competency knowledge graph; the subgraph weight generation module is used to generate scenario query conditions, extract conditional subgraphs, and generate dynamic evidence weights; the vector representation module is used to call the heterogeneous graph Transformer to generate scenario-based employee competency vectors and job competency benchmark vectors; the calibration and evaluation module is used to generate competency scores, competency gap sets, job suitability, and confidence intervals; and the career planning module is used to generate career development paths or supplementary certification instructions, training instructions, or instructions not to recommend at this time.

[0045] The beneficial effects of this invention are:

[0046] 1. By constructing a time-series heterogeneous capability knowledge graph from employees, positions, tasks, equipment, risk scenarios, capability tags, courses, certificates, and business events, and writing timestamps and dynamic evidence weights to evidence edges, it is possible to unify the organization of capability evidence in multiple business systems and reduce the impact of inconsistent data standards and different update cycles on capability assessment results.

[0047] 2. By extracting conditional subgraphs based on job position, task, equipment, and risk level, and introducing relational encoding, time encoding, scenario query encoding, and dynamic evidence weight bias in the attention calculation of the heterogeneous graph Transformer, it is possible to match employee capability vectors and job capability baseline vectors with specific work scenarios, thereby improving the relevance of capability scores, capability shortcomings, and job suitability outputs.

[0048] 3. By adopting conformal prediction to generate confidence intervals for capability scores and job suitability, and incorporating confidence interval width, job suitability threshold, job safety threshold, training resources, and certification cycle into multi-objective programming, it can output supplementary certification instructions, training instructions, or temporary non-recommendation instructions when the assessment confidence is insufficient or the risk constraints are not met, thereby improving the reliability of career development path generation and the adaptability of safety control. Attached Figure Description

[0049] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0050] Figure 1 is a flowchart of a method for employee competency assessment and career development planning of power grid enterprises.

[0051] Figure 2 is a flowchart of step S2 of the present invention for generating a conditional subgraph and dynamic evidence weights. DETAILED IMPLEMENTATION

[0052] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0053] referring to Figures 1-2 , a method for employee competency assessment and career development planning of power grid enterprises, comprising: S1, collecting employee competency evidence data from training, examination, qualification, operation record, defect handling and safety management and control systems, performing unified mapping on data source, time, employee, post, task, equipment, risk level and competency label, and generating a time-series heterogeneous competency knowledge graph including business event nodes and certificate holding edges; S2, generating scene query conditions including post, task, equipment and risk level for the employee to be assessed and the target post, extracting a conditional subgraph, and generating dynamic evidence weights of evidence edges; S3, inputting the conditional subgraph into a trained heterogeneous graph Transformer, forming an attention bias according to relation encoding, time encoding, scene query encoding and dynamic evidence weights, and outputting a scene-based employee competency vector and a post competency reference vector; S4, performing dimension alignment and difference calculation on the two vectors to generate a competency score, a competency shortboard set, post adaptability and a confidence interval thereof; S5, inputting the competency shortboard set, post adaptability, confidence interval, training resources, qualification obtaining period and risk level into a multi-objective programming model, outputting a career development path when the post adaptability reaches a post adaptability threshold, the width of the confidence interval does not exceed a confidence interval threshold, and the risk level is not higher than a post safety threshold, otherwise outputting a qualification supplement instruction and / or a training instruction and / or a temporary non-recommendation instruction.

[0054] In this specific implementation, S1 comprises:

[0055] The business event node is a superordinate node type of operation events, at least including training event, examination event, on-site operation event, defect handling event and safety management and control event; the risk level represents that the risk increases from low to high in the order of R1<R2<..., the post safety threshold represents the highest acceptable risk level of the target post, and it is determined that the post safety requirement is satisfied when the risk level is not higher than the post safety threshold;

[0056] In this specific implementation, the competency assessment platform acts as the implementing entity. At the beginning of each assessment batch, it retrieves employee competency evidence data from the training management system, examination system, qualification and certification system, work record system, defect handling system, and safety control system. Each piece of raw data is then compiled into an evidence record. This evidence record includes at least the data source identifier, generation time, update time, business object identifier, event result, risk level, employee original identifier, job original identifier, task original identifier, equipment original identifier, and competency tag original identifier. The platform calculates the evidence primary key using the data source identifier, business object identifier, generation time, and employee original identifier. When the same If a record already exists, retain the record with the newest update time and write the old record into the version chain. When a key field is missing, write the record into the verification queue instead of writing it into the graph as valid evidence.

[0057] The capability assessment platform reads a pre-defined field dictionary to uniformly map data sources, time, employees, positions, tasks, equipment, risk levels, and capability tags. This pre-defined field dictionary is generated before deployment from HR master data, job capability standards, equipment ledgers, risk control lists, and course certificate catalogs. Each record in the field dictionary includes the original system field name, original system value, unified object type, unified identifier, effective start and end times, and dictionary version. The mapping rule is denoted as ,in The values ​​are taken from the original system. For object type, To ensure uniform identification, if the same original system value maps to multiple unified identifiers, the record with the latest version number and whose effective start and end times cover the evidence generation time will be used to determine the identifier. If it is still impossible to uniquely determine the evidence, mark the evidence record as pending manual confirmation.

[0058] After completing the unified mapping, the capability assessment platform establishes node write buffers and edge write buffers. The node write buffers are configured according to the unified identifier, node type, and dictionary version. Perform idempotent merging, writing to the buffer according to the evidence primary key. Generate edge keys from start node, end node, and relation type. When the same When dealing with multiple event results, the platform does not overwrite the old edges, but instead saves the result sequence, source chain, and valid state in the edge attributes, so that subsequent steps can distinguish between continuously valid evidence, expired evidence, and conflicting evidence.

[0059] The capability assessment platform uses unified employee identifiers, unified job identifiers, unified task identifiers, unified equipment identifiers, risk scenario identifiers, capability tag identifiers, course identifiers, certificate identifiers, and business event identifiers as node sets, and job title, participation, associated equipment, corresponding risk, demonstrated capability, required capability, completed course, and held certificate as edge type sets. It constructs a time-series heterogeneous capability knowledge graph, where the node type represents the node, the relationship type represents the edge, and the evidence generation time represents the evidence generation time. The edge attributes include event result, risk level, data source identifier, evidence primary key, and update timestamp. The graph forms a time-series index based on event time rather than data entry time.

[0060] For training and examination-related evidence, the capability assessment platform establishes a course completion edge from employee nodes to course nodes or examination business event nodes, and connects the course nodes or examination business event nodes to capability tag nodes via capability proof edges. For qualification-related evidence, the capability assessment platform establishes a certificate holding edge from employee nodes to certificate nodes, and connects the certificate nodes to capability tag nodes via capability proof edges. For on-site operations, defect handling, and safety management-related evidence, the capability assessment platform establishes an edge structure consisting of employee nodes participating in business event nodes, business event nodes associated with task nodes and equipment nodes, business event nodes corresponding to risk scenario nodes, and business event nodes proving capability tag nodes. If the event result indicates failure, defect not being closed, or safety violation, a negative result marker is written in the corresponding capability proof edge; otherwise, a positive result marker is written.

[0061] The preset edge type dictionary maps the records of certificate holding, renewal approval, and certificate validity in the qualification and certification system to certificate holding edges, the records of training completion and exam passing to course completion edges, and the support of courses, certificates, or business events for capability tags to capability proof edges. This edge type dictionary is loaded and cached together with the preset data source level mapping table at the beginning of the graph construction batch. The data source level is a discrete level, and the data source reliability value is only used when S2 calculates the dynamic evidence weight.

[0062] For the mapping between risk levels and capability tags, the capability assessment platform calls the risk level table and capability tag hierarchy table in the job capability standards. The risk level table includes the original risk name, unified risk level, applicable equipment category and job safety threshold reference. The capability tag hierarchy table includes capability tag identifier, parent tag, job family, evidence type and minimum number of evidence. If the original evidence only contains the parent capability tag, the platform expands it to the corresponding sub-tag according to job family and task type and writes an inheritance mark in the evidence edge. The inheritance mark reduces the scenario relevance of the evidence edge when S2 calculates the dynamic evidence weight.

[0063] The capability assessment platform performs deduplication on duplicate evidence for the same employee, the same capability tag, the same business object identifier, and the same time range of occurrence. In this specific implementation, the deduplication window is set to 24 hours. The retention rule is to first compare the reliability level of the data source, and then compare the update timestamp. When the reliability level and update timestamp are the same, the record whose event result is closer to the actual work performance is retained. After deduplication, the list of data source identifiers of the merged records is still retained as the evidence source chain. The source chain is used for subsequent cross-system consistency coefficient calculation without changing the original event result.

[0064] Before writing the graph, the capability assessment platform performs time validity checks on the edge timestamps. If the generated time is later than the update time or the assessment batch time, the evidence is recorded in the abnormal time schedule. If the generated time falls outside the certificate validity period, course completion cycle, or assignment plan window, a time out-of-bounds flag is written in the edge attribute. The time out-of-bounds flag does not delete the evidence edge, but it causes the edge to enter the low-weight branch when S2 calculates the time decay coefficient.

[0065] The time-series heterogeneous capability knowledge graph is written into a graph database and an assessment feature library. The graph database establishes a combined index based on employee unified identifiers, job unified identifiers, capability tag identifiers, risk scenario identifiers, and timestamps. The assessment feature library stores evidence records, node tables, edge tables, and dictionary versions. The platform outputs the graph version after completing the batch build, showing the reference relationships between them. Valid evidence record set And a list of data to be verified, the version of the map It serves as the input for generating scenario query conditions, extracting conditional subgraphs, and calculating dynamic evidence weights in S2.

[0066] In this specific embodiment, S2 includes:

[0067] The subgraph weight generation module reads the graph version, the unified identifier of the employee to be evaluated, and the unified identifier of the target position. It generates scenario query conditions containing position, task, equipment, and risk level from the job competency standards and the current work plan. The scenario query conditions include the target position identifier, the set of job requirement capability tags, the set of target tasks, the set of target equipment, the minimum job safety threshold, the set of risk levels, and the evaluation time. The job safety threshold is derived from the job safety threshold table and is used to compare with the evidence risk level. If the evidence risk level is not higher than the job safety threshold, it is determined that the job safety requirements are met. When there is no specific equipment for the target position in the current work plan, the module reads the equipment category in the job standard as the equipment query field and writes the generalized equipment tag.

[0068] The subgraph weight generation module extracts conditional subgraphs based on search criteria. The extraction rules are to retain directly connected evidence edges, capability label nodes connected to the required capability edges of the target position, business event nodes that match the target task or target equipment, and course nodes and certificate nodes that are connected to the nodes within two hops. If the risk scenario, task, and equipment of an evidence edge have no matching items, the evidence edge is removed. If a certain capability label of a position has no evidence edges after extraction, the capability label node is retained and a missing evidence mark is written in the subgraph.

[0069] For non-scenario-based evidence such as training, examinations and qualifications, the subgraph weight generation module does not require it to have task, equipment and risk fields at the same time. When the capability label associated with the edge of this type of evidence matches the set of capability labels required by the target position, the evidence edge is retained in the conditional subgraph and its risk scenario association degree is set to the default value of low scenario of the job family. The three types of reasons in the log are removed: missing distinguishing fields, mismatched capability labels and mismatched scenario fields.

[0070] Conditional subgraph extraction is performed using graph indexing and candidate edge filtering algorithms. The module first uses a unified employee identifier. Uniform job identification Retrieve candidate edge sets using the capability label set as the index key. Then calculate the matching predicate for each candidate edge. ,in Indicates evidence linking employees and Consistent This indicates that the evidence edge's association capability label belongs to the target job requirement set; the edge enters the target job when the predicate is true. When the predicate is false, the edge is written to the pruning log and the reason for pruning is retained;

[0071] for Every piece of evidence in The subgraph weight generation module calculates the dynamic evidence weights. The dynamic evidence weight is determined by the reliability of the data source. Time decay coefficient Cross-system consistency coefficient Correlation with risk scenarios The result is obtained by normalizing and weighting, and the calculation formula is: Each component was cut to The denominator is fixed as the sum of the four weight coefficients. If a component cannot be calculated due to missing fields, then the component takes the lowest confidence default value of the job family and a missing component marker is written into the edge attribute. The weights are derived from the job family weight configuration table, which uses job category and risk level as the search key and uses the same version in this batch. ;

[0072] Data source reliability The data source level is determined by a preset data source level mapping table, which includes a data source identifier, data source level, reliability value, applicable business object, version number, and effective time. An example record is one where the data source identifier is the qualification and certification system, the data source level is Level 1, the reliability value is 0.95, the applicable business object is the certificate node, and the version number is... If the data source identifier of the evidence edge does not match the mapping table, then... Set the edge to the lowest historical reliability for the same job family and mark it as a source pending confirmation;

[0073] Time decay coefficient According to the assessment time With the moment the evidence was generated The interval and the validity period parameter corresponding to the position are determined, and the calculation formula is: ,in Measured by day, For the target position In the job validity period parameter table, if the evidence is a permanent qualification certificate and the certificate has not expired, then the remaining validity days of the certificate shall be used. The maximum value in the formula is used as the denominator for attenuation. If the certificate has expired, then... Set to 0 and retain the evidence edge but do not participate in the positive ability accumulation;

[0074] Cross-system consistency coefficient The ratio of consistent evidence results across different systems within the same employee, competency label, and risk scenario to the number of comparison participants is determined using the following formula: ,in To be on the edge of the evidence The amount of historical evidence that aligns with the outcome of the event To determine the number of historical evidences to be compared, the smoothing parameter... In this specific implementation, the value is 1, when hour The value is set to 0.5 and the initial evidence marker is written, which is used for confidence interval width correction in S4;

[0075] Risk scenario relevance Based on the task, equipment, risk level, and scenario search criteria of the evidence event. The number of matching terms is determined, and the calculation formula is: ,in , and These are the matching indicators for task, device, and risk level, respectively. A match is 1 and a non-match is 0. If the device field uses a generalized device tag, then the device category matching... If S1 writes the capability tag inheritance flag, then... Multiply by inheritance reduction factor , The values ​​are given by the capability tag hierarchy table and range from [value range missing]. The value is used to reduce the weight of evidence that migrates across devices or levels without directly deleting the evidence edge;

[0076] The subgraph weight generation module performs a normalization consistency check before writing the weights. The check includes... , , and Value range, participating denominator, weight configuration version and scenario query conditions Version, when or If the edge is still out of bounds, it will be recalculated based on the original components in the edge attributes. The edge is clipped to the nearest boundary and the reason for the out-of-bounds error is written into the weight audit field, which is used to identify low-quality evidence edges when attention bias is formed in S3.

[0077] When a new evidence record is added to the graph, the subgraph weight generation module retrieves historical evidence records for the same employee, the same capability tag, and the same risk scenario based on the business object identifier of the new evidence record. The retrieval scope includes valid evidence prior to the time the new record was generated and concurrent evidence within the same task window as the new record. The module updates the graph with the event results of the new record. and Update with the time when the new record is created. And maintain the reliability of the data source. Correlation with risk scenarios The source version remains unchanged;

[0078] The subgraph weight generation module will correct the... And recalculated and , Renormalization yields the updated Renormalization still uses As a fixed denominator, if a new record causes a conflict between positive and negative evidence under a certain capability label, then a conflict flag is written for the relevant evidence edges of the same capability label, and the cross-system consistency coefficient is reduced. The updated Weighted Version The update time and conflict marker are written back to the edge attributes of the corresponding evidence edge;

[0079] The subgraph weight generation module ultimately outputs a conditional subgraph object containing a node table, an edge table, an evidence edge weight table, and a pruning log. The fields of the evidence edge weight table include the evidence primary key. Side keys , , , , , The conditional subgraph includes fields for missing component markers, conflict markers, and weight auditing. Together with dynamic evidence weights This serves as the input for S3 to form the attention bias and output the capability vector.

[0080] In this specific embodiment, S3 includes:

[0081] The vector representation module reads the conditional subgraph, dynamic evidence weights, scenario query conditions, and trained heterogeneous graph Transformer model output by S2. The input nodes of the model include employees, positions, tasks, equipment, risk scenarios, capability labels, courses, certificates, and business events. The input edges include holding a position, participating, associated equipment, corresponding risks, proving capabilities, requiring capabilities, completing courses, and holding certificates. The module constructs the model input sequence for each node and its incoming edges according to node type, relationship type, time interval, and scenario query fields.

[0082] The heterogeneous graph Transformer is obtained through self-supervised training before deployment. The training samples are generated by the historical time-series heterogeneous capability knowledge graph according to the evaluation month and job family slices. Each training sample includes a historical conditional subgraph, evidence edge weights, real subsequent business event results and job capability label consistency markers. The training tasks include occlusion node type recovery, occlusion relationship type recovery, future business event result prediction and capability label consistency discrimination. After training, the model parameters are fixed and the model version is recorded. Only forward inference is performed in the online evaluation phase.

[0083] When training samples are generated, the vector representation module extracts positive and control samples from the historical graph by employee, job family and evaluation month. Positive samples are subgraphs where employees have completed tasks in the target job or job family and the subsequent event results have been closed. Control samples are subgraphs with similar jobs but missing competency labels or unclosed risk events. The sample record saves the sample identifier, graph version, job family, evaluation month, set of occluded nodes, set of occluded edges, future business event labels and sample weights. The sample weights are jointly determined by the average weight of historical evidence edges and the job risk level.

[0084] In this specific implementation, the heterogeneous graph Transformer model structure includes a node type embedding layer, a relation type embedding layer, a time interval embedding layer, a scene query embedding layer, and a multi-layer heterogeneous attention encoder. Initial node features are obtained by concatenating node type, capability label level, job requirement strength, event result label, and missing evidence label. Initial relation features are obtained by encoding relation type, evidence edge weight, data source identifier, and risk level. The number of encoder layers, attention heads, and hidden dimensions are given by the training configuration table and vary with the model version. Save together;

[0085] Event outcome labels enter the heterogeneous graph Transformer through independent channels. Positive and negative outcome labels participate in the generation of positive and negative evidence value vectors, respectively. During attention aggregation, dynamic evidence weights only represent the reliability of evidence and the strength of scene association, without changing the direction of event outcome. Highly reliable negative evidence is aggregated into the negative capability evidence channel to represent insufficient capability, risk exposure, or a state pending review, and is not interpreted as positive capability enhancement.

[0086] The total loss during the training phase is calculated as follows: , where is the cross-entropy loss of the occlusion node type, is the cross-entropy loss of the occlusion relationship type, is the prediction loss of future business event results, is the loss of capability label consistency discrimination, is the evidence weight bias regularization term, is the bias mean under the same relationship type, and the weights are derived from the training configuration table and training stops when the job suitability error in the validation set no longer decreases.

[0087] During online inference, the vector representation module embeds the node type. Relational type embedding Time interval embedding and scene query embedding After concatenation, the mapping is into a query vector, a key vector, and a value vector, and the calculation formula is: , , ,in , and For the model Fixed parameters in the time interval embedding From the evaluation time with edge timestamp Obtained by dividing into buckets;

[0088] The vector representation module will dynamically assign evidence weights. Transformed into attention bias terms via monotonic mapping table The monotonic mapping table uses weight intervals as keys and stores bias values, applicable relation types, versions, and effective times. The mapping satisfies... hour In this specific embodiment, according to Calculation, where Obtained from the validation set, if the evidence edge lacks the dynamic evidence weights that have been written back, then low-confidence dynamic evidence weights are first generated according to the default values ​​of the four components of S2. Then converted through a monotonic mapping table And write the low-weight source tag;

[0089] The monotonic mapping table is generated by validation set calibration before model deployment. The calibration process uses the evidence edge weight interval as the bucket key, and counts the prediction error of future business events and the consistency error of capability labels in each bucket. If the error of a high-weight bucket is lower than that of a low-weight bucket, the corresponding bias ladder is maintained. If an inverse relationship occurs, adjacent buckets are merged until monotonicity is satisfied. The calibrated table records the lower bound of the weight, the upper bound of the weight, the bias value, the relationship type, the model version, and the number of samples. Buckets with a sample number lower than the calibration sample threshold use the bias value of the adjacent bucket. The calibration sample threshold is derived from the model deployment configuration and is compared with the number of samples in the bucket to be no less than the required number.

[0090] In each attention head, the model adds the attention bias term to the dot product of the query vector and key vector corresponding to the same evidence edge, and the attention weights are adjusted accordingly. Calculation, where For single-head dimension, For relational encoding bias, For time coding offset, The attention bias is formed by dynamic evidence weights, and and Monotonically homogeneous, evidence edges with higher weights receive higher attention contributions when aggregated in the same neighborhood;

[0091] The model aggregates the value vectors of neighboring nodes according to node type, and the aggregation rule is as follows:

[0092] ;

[0093] in For nodes The type is The neighborhood group, For node type The corresponding projection matrix, As a non-linear mapping, if a certain capability label node has no evidence neighbors, the representation of the label node is jointly generated by the job requirement capability edge and the missing evidence embedding, and the missing label is retained in the output.

[0094] After the neighbor aggregation is completed, the vector representation module performs job scenario pooling on the capability tag nodes. The pooling rule is to read the corresponding capability tag node representation according to the set of capability tags required for the job, and calculate the tag-level representation according to the job requirement intensity and risk level weight. If a capability tag comes from course, certificate and business event evidence at the same time, the module retains the attention contribution ratio of positive evidence, negative evidence and course, certificate and business event three types of sources respectively. The contribution ratio is written into the vector result table so that S4 can identify the risk of single source evidence when generating the confidence interval.

[0095] The vector representation module reads the contextualized employee capability vector from the final layer representation of employee nodes, capability tag nodes, and target position nodes. and job competency benchmark vector , Arranged in order of the target job requirements and integrating employees' skills and abilities. The positive and negative evidence in the text indicates that... The vectors are generated based on the competency requirements and risk scenarios in the job standards. Both vectors are written into a vector result table, which includes employee identifier, target job identifier, competency tag order, and model version. Atlas version Scenario query conditions and missing evidence markers;

[0096] When outliers are present in the capability labels output by the model, the vector representation module calculates the Mahalanobis distance based on the historical vector distribution of the same job family. ,like If the abnormal output threshold is exceeded, the recalibration of the capability label dimension is triggered and it is marked as abnormal output. The abnormal output threshold and recalibration parameters are derived from the model output calibration table. The module uses the employee's most recent effective evaluation vector of the same job family for smoothing. The abnormal label, the value before smoothing, and the value after smoothing are all written into the vector result table to avoid abnormal dimensions being misjudged as real capability shortcomings in S4.

[0097] when The number of valid evidence edges in the model is lower than the minimum evidence threshold for the job family or the model version. When the target job family is not applicable, the vector representation module does not delete the evaluation task, but instead... Write a low-confidence flag and call the most recently applicable model version for inference, while writing the triggering reason into the vector result table. If there is no most recently applicable model version, then use the job competency benchmark vector. The missing evidence is embedded to generate a conservative employee capability vector, and the task is forcibly marked as requiring supplementary evidence. This contextualized employee capability vector... Job competency benchmark vector And low-confidence markers are used as inputs for S4 to perform dimension alignment, difference calculation and confidence interval generation.

[0098] In this specific embodiment, S4 includes:

[0099] The calibration and assessment module reads the scenario-based employee capability vector output by S3. Job competency benchmark vector The order of competency labels, missing evidence markers, and low confidence markers are used as the job competency baseline vector. The dimensional order is used as the standard dimensional sequence. ,like Lack If a certain ability label dimension is selected, then the dimension is populated with the historical mean of that job family, and the corresponding ability label is added to the missing evidence count. Includes those that do not belong The additional dimensions will be saved as reference ability fields and will not be included in the job suitability calculation.

[0100] The calibration and evaluation module normalizes the two vectors based on the job family normalization parameter table. The normalization rule is as follows: , ,in For the ability tag dimension, and Samples of qualified employees from the same job category. To prevent smooth terms with a denominator of 0, the normalized terms are... and All were cut into the designated range of job families. An exception vector flag is written when the clipping is triggered;

[0101] The calibration evaluation module calculates the dimension-wise difference vector. ,in Indicates employee's competency label The performance is below the job competency benchmark. This represents the difference vector between the employee's performance and the benchmark performance for their position. Together with competency labels, evidence missing markers, and risk scenario markers, these are written into a difference detail table, which serves as a unified input for competency scores, competency deficit sets, and job suitability.

[0102] Competency scores are generated from a scoring mapping table, which uses job family, competency label, and risk level as keys. The table stores the shortcoming threshold, scoring slope, and minimum acceptable score. The calibration assessment module is configured accordingly. Computing ability tag Ability scores, among which Ability labels in the rating mapping table The upper limit of the normalized difference, The range of values ​​is , The larger The lower the risk level, the better. If the capability label corresponds to a high-risk task, then the lower the risk level should be. To improve the sensitivity of identifying weaknesses;

[0103] Collection of capability shortcomings according to Generate, where Capability Tags The minimum ability score threshold, The threshold for the shortcoming difference is derived from the scoring mapping table. If a certain ability label has a missing evidence marker and In Within a range of 5 points above and below, the capability tag is simultaneously written into the "shortcomings to be certified" field. The "shortcomings to be certified" field is used to output the data source identifier and capability tag to be supplemented when S5 generates the certification instruction.

[0104] Job fit is generated by a fit mapping model. In this specific implementation, the fit mapping model adopts a logistic regression model calibrated with historical evaluation samples, and the input is a one-dimensional difference vector. The risk level coding and the proportion of missing evidence are calculated using the following formula:

[0105] ;

[0106] in The value range is for job suitability. , For risk level normalization coding, The proportion of missing evidence, , and From the fitness mapping model parameter table, The larger the value, the higher the job fit.

[0107] The training samples for the fit mapping model are constructed from historical evaluation samples. Positive class labels are defined as employees who have no high-risk violations, pass key operations on the first attempt, and have a defect recurrence rate lower than the job family threshold within 90 days of taking up the target position or a similar position. Negative class labels are defined as employees who have high-risk violations, fail key operations, or have a defect recurrence rate not lower than the job family threshold. During training, the time is divided according to the evaluation month. Early samples are used for training, and samples from later months are used for validation. Class imbalance is corrected by positive and negative class weights. The intercept, coefficients of each capability label, risk level coefficient, missing evidence ratio coefficient, sample window, and model version obtained during training are all saved to the fit mapping model parameter table.

[0108] The confidence interval is obtained by constructing the non-compliance score from the predicted residuals of historical assessment samples. The calibration assessment module calculates the capability score non-compliance on the calibration set of the same job family. Job fit and mismatch In target coverage The following are the quantiles of the non-compliance score distribution. and ,in and In this specific implementation, the target coverage rate is determined by the job risk level, with a target coverage rate of not less than 0.95 used for high-risk jobs;

[0109] The observed true values ​​of the competence scores are obtained by mapping the work closed-loop quality, safety violation rate, defect recurrence rate, and test retest scores within the subsequent observation window. The observed labels of job suitability are obtained by the results of subsequent job appointments or trial positions in the same job family. The prediction residual is defined as the difference between the model's predicted competence score or job suitability and the corresponding observed index mapping score or observed suitability label. Therefore, the sample source, observation window, and mapping rules of the non-compliance scores are all written into the confidence interval parameter table along with the calibration set version.

[0110] The calibration evaluation module is based on and Generate confidence intervals for competency scores and job suitability, where the confidence interval for competency scores is... The job suitability confidence interval is: If S3 writes a low-confidence marker or S2 writes an initial evidence marker, then in the corresponding dimension... or Multiply by confidence amplification factor The calibrated from historically low evidence sample coverage;

[0111] The calibration evaluation module outputs evaluation result records. The evaluation results record includes employee identification, target position identification, and competency tag scores. Dimensional difference vector A collection of capability shortcomings Job suitability The assessment results include: confidence interval for ability score, confidence interval for job suitability, confidence interval width, number of missing evidence, risk level, and version of the scoring mapping table. The results are written into the evaluation results database and used as the basis for the S5 input multi-objective programming model to generate career development paths and related instructions;

[0112] The number of missing evidence is obtained by counting the missing tags in the set of job requirement capability tags, where missing ordinary capability tags are counted as 1 and missing capability tags corresponding to high-risk job tasks are counted as 2. The data source identifier to be supplemented is traced back in three levels: evidence type, business system and key field. The certificate system, examination system or security control system identifier that can directly prove the capability tag is output first, and then the related course or business event field is output.

[0113] In this specific embodiment, S5 includes:

[0114] The career planning module reads the evaluation results recorded by S4. The training resource catalog, certificate catalog, job task list, certification cycle table, and job safety threshold table are compiled, and are organized by competency gaps. Job suitability Confidential interval width and risk level As part of the planning stage, each record in the training resource catalog includes a course identifier, covered competency tags, supplementary scope, training duration, course schedule, applicable job families, and resource capacity. Each record in the certificate catalog includes a certificate identifier, associated competency tags, certification period, validity period, and prerequisite courses.

[0115] The career planning module will address any skill gaps. The competency tags in the data establish an reachability relationship with courses, certificates, job tasks, and certification periods. If the set of competency tags covered by a course or certificate... and The intersection of the values ​​is not empty and the applicable job families include the target job. Then candidate development actions are generated. Candidate development actions The fields include action type, course identifier, certificate identifier, associated job task, and set of covered ability tags. Expected start time, expected completion time, and training duration (calculated in days) Number of days for evidence collection Resource capacity flags and risk constraint fields;

[0116] Reachability relationships are constructed as action graphs starting with ability tags. ,in For the set of candidate development actions, For the reachable edges from competency tags to courses, courses to certificates, and certificates to job tasks, edge attributes include prerequisite dependencies, resource capacity, earliest start date, and risk scope. If a competency tag can only be supplemented by job practical tasks without being covered by courses or certificates, the module generates practical follow-up actions and sets the completion conditions of the action as the simultaneous existence of target task closed-loop records and safety control confirmation records.

[0117] For each candidate development action The career planning module inputs information in a planning state. A collection of capability shortcomings Confidential interval width and risk level Computational capacity supplement for the computation window:

[0118] ;

[0119] in Capability Tags The standardization gap Competency tags for courses or certificates The calibration compensation range, The results were obtained by fitting the training effectiveness evaluation samples and the post-certification performance samples. If the action... If the preconditions are not met, the action is marked as inactive and does not enter the objective function. The larger the value, the greater the contribution of the candidate development action to addressing the set of shortcomings;

[0120] The career planning module simultaneously calculates the training conversion days, certification time days, and risk constraint satisfaction flag. Here, is the class hours in the training resource catalog, is the daily available training hours in the job family planning weight table, is the waiting days for the course to start, which comes from the certification cycle table and is in days. The risk constraint satisfaction flag indicates that the risk level of the job task covered after the action is completed is not higher than the job safety threshold and the required certificate is valid. If the candidate development action lacks available course capacity or the certificate prerequisite courses are not met, the candidate action is retained in the candidate table but marked as prohibited from selection.

[0121] When the job fit reaches the job fit threshold, the confidence interval width does not exceed the confidence interval threshold, and the risk level is not higher than the job safety threshold, the career planning module starts a multi-objective programming model to select candidate development action sequences. The decision variable represents whether to select candidate actions. The objective function is to maximize, where is the positive weight of the ability supplementation, and is the penalty weight of the training conversion days and the certification time days. The weights come from the job family planning weight table or the upper limit of the historical action samples of the same job family. The three weights come from the job family planning weight table and the sum is 1.

[0122] The multi-objective programming model sets resource capacity, prerequisite courses, certification deadlines, and risk constraints. Constraints include that course capacity cannot exceed the available quota during the same time period, certification actions must be taken after prerequisite course actions, and the total completion time cannot exceed the job transfer window. The selected actions all satisfy Furthermore, the competency labels corresponding to high-risk job tasks must be covered by at least one course, certificate, or hands-on training activity;

[0123] When the integer programming problem has no feasible solution within the 30-second time limit set in this specific implementation, the career planning module first identifies the set of infeasible constraints. The set of infeasible constraints includes insufficient resource capacity, lack of prerequisite courses, and exceeding the certification period. Risk constraints Actions not covered by the high-risk capability label, if If the resource capacity is insufficient, the module will be configured as follows: Perform a greedy selection from high to low and write the alternative class opening time into the path. If the risk constraints are not met or the high-risk capability label does not cover the action, then the recommended development path will be stopped and the branch of generating instructions that are not recommended will be entered.

[0124] The career planning module generates a career development path based on the prerequisites and expected completion time of the selected actions. Each node in the path includes an action sequence number, course identifier, certificate identifier, associated job task, supplementary competency label, estimated start time, estimated completion time, and estimated job fit after completion. The estimated job fit after completion is recalculated using the S4 fit mapping model, and the standardized gaps in the supplemented competency labels are adjusted accordingly. If multiple candidate actions contribute the same amount to the objective function, the action with more high-risk job task capability labels will be selected first. If they are still the same, the action with shorter certification time will be selected.

[0125] The career planning module performs a path validity replay after path generation, updating the set of skill gaps, certificate validity status, course completion status, and job task risk constraints item by item according to the action number. If the replay result shows that the last action has been completed... Still below the job suitability threshold If there are still high-risk capability labels that are not covered, then mark the path as an ineffective path and output supplementary action requirements or instructions that are not recommended for the time being, so as to avoid generating development paths that cannot meet the safety requirements of the position just because the local action scores are ranked first.

[0126] The career planning module first calculates the objective triggering conditions without the preconditions of the main instruction priority: when the confidence interval width exceeds the confidence interval threshold and the number of missing evidence reaches the supplementary evidence threshold, it is determined that the supplementary evidence condition is met. The supplementary evidence instruction includes the data source identifier to be supplemented, the ability tag, the missing job task, the target completion date and the triggering reason. The data source identifier to be supplemented is obtained by tracing back the evidence source chain of S1 from the shortcoming field to be supplemented in S4. If the number of missing evidence does not reach the threshold, only the observation mark is written in the planning result record.

[0127] When the confidence interval width does not exceed the confidence interval threshold and the capability shortcoming set is not empty, the training conditions are deemed met. The training instruction includes a course identifier, a certificate identifier, and a completion deadline. The completion deadline is calculated as follows: Confirmed, among which For the assessment time, , and The units are all days. If the candidate development action sequence is empty, the available courses that cover the most shortcomings will be used as the bottom-line training instructions and the uncovered capability tags will be written into the subsequent supplementary fields.

[0128] When the risk level is higher than the job safety threshold or the job suitability is lower than the job suitability threshold, the conditions for temporary non-recommendation are met. The content of the temporary non-recommendation instruction includes the unmet risk level, job task, competency label, current job suitability, and minimum qualification conditions. If the conditions for supplementary certification and / or training are met at the same time, the temporary non-recommendation instruction will be marked as the main instruction in the planning results record, and the supplementary certification instruction and / or training instruction will be marked as the auxiliary instructions for lifting the temporary non-recommendation status.

[0129] When the conditions for temporary non-recommendation are not met but the conditions for supplementary certification are met, the career planning module will designate the supplementary certification instruction as the primary instruction; when the conditions for temporary non-recommendation and supplementary certification are not met but the conditions for training are met, the training instruction will be designated as the primary instruction; when both the conditions for supplementary certification and training are met but the conditions for temporary non-recommendation are not met, the supplementary certification instruction will be designated as the primary instruction and the training instruction as the secondary instruction, in the order of priority of credible evidence over addressing shortcomings. Both the primary and secondary instructions will be written into the instruction queue and bound to the same assessment result record identifier.

[0130] The career planning module ultimately outputs a career development path, a certification supplement instruction, a training instruction, or a temporary non-recommendation instruction, as well as a planning result record. When the job suitability reaches the job suitability threshold, the confidence interval width does not exceed the confidence interval threshold, the risk level is not higher than the job safety threshold, and the path validity playback passes, the career development path field in the planning result record is written with the selected action sequence. When a certification supplement instruction, training instruction, or temporary non-recommendation instruction is triggered, the career development path field is empty or written with a pending generation status. The planning result record saves the evaluation result record identifier, candidate action set, objective function value, constraint satisfaction status, threshold version, instruction type, instruction details, and effective time. The record is written into the employee's career development file and fed back to the training management system, qualification and certification system, and job appointment approval process.

[0131] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0132] This invention uses a time-series heterogeneous capability knowledge graph to link employee capability evidence, job requirements, equipment objects, and risk scenarios across multiple business systems. This enables previously scattered training, examination, qualification, operation, and safety management data to form a computable chain of evidence around the same employee, capability tag, and job task. Furthermore, by using conditional subgraphs and heterogeneous graph Transformers to output scenario-based capability representations, the capability assessment results can correspond to specific job tasks, equipment types, and risk levels, thereby solving the problem that fixed indicators and manual scoring cannot reflect the true differences in capabilities.

[0133] This invention further incorporates data source reliability, time decay, cross-system consistency, and risk scenario relevance into dynamic evidence weights, and uses these weights as attention biases in graph Transformer calculations. Simultaneously, at the output end, it distinguishes between supplementary evidence, training, and temporary non-recommendation results through credible interval constraints and multi-objective programming. This enables the assessment model to not only output capability scores and job suitability, but also provide executable development paths and handling instructions based on evidence sufficiency and security risks, thereby better achieving the timeliness, scenario adaptability, and security reliability of capability assessment.

Claims

1. A method for assessing the competence and career development planning of employees in power grid enterprises, characterized in that, include: S1. Collect employee capability evidence data from training, examination, qualification, work records, defect handling and safety management systems, and perform unified mapping of data source, time, employee, position, task, equipment, risk level and capability tag to generate a time-series heterogeneous capability knowledge graph containing business event nodes and certificate holding edges. S2. Generate scenario query conditions containing job, task, equipment and risk level for the employees to be evaluated and the target positions, extract conditional subgraphs, and generate dynamic evidence weights for evidence edges. The dynamic evidence weight is obtained by normalizing and weighting the data source reliability, time decay coefficient, cross-system consistency coefficient, and risk scenario relevance. The reliability of the data source is determined by a preset data source level mapping table; The time decay coefficient is determined based on the interval between the assessment time and the evidence generation time, as well as the validity period parameter corresponding to the position; the cross-system consistency coefficient is determined based on the ratio of the number of consistent evidence results for the same capability label in different systems to the number of comparisons. The risk scenario relevance is determined based on the number of matching items between the task, device, and risk level of the evidence event and the scenario query conditions; dynamic evidence weight is used as the edge attribute of the evidence edge in the conditional subgraph. The dynamic evidence weight is updated when a new evidence record is added. Specifically, it includes: retrieving historical evidence records under the same employee, the same capability tag, and the same risk scenario based on the business object identifier of the new evidence record; correcting the cross-system consistency coefficient based on the consistency between the results of the new evidence record and the historical evidence record; recalculating the time decay coefficient based on the generation time of the new evidence record; and renormalizing the corrected cross-system consistency coefficient and the recalculated time decay coefficient with the data source reliability and the correlation of the risk scenario to obtain the updated dynamic evidence weight, and writing it back to the corresponding evidence edge. S3. Input the conditional subgraph into the trained heterogeneous graph Transformer, form an attention bias based on relation encoding, time encoding, scenario query encoding, and dynamic evidence weights, and output scenario-based employee capability vectors and job capability benchmark vectors; concatenate node type embedding, relation type embedding, time interval embedding, and scenario query embedding and map them into query vectors, key vectors, and value vectors; convert the dynamic evidence weights into attention bias terms through a monotonic mapping table, and add the attention bias terms to the dot product of the query vector and key vector corresponding to the same evidence edge; aggregate the value vectors of neighboring nodes according to node type and distinguish between positive and negative evidence, and output scenario-based employee capability vectors and job capability benchmark vectors. S4. Perform dimension alignment and difference calculation on the two vectors to generate capability scores, capability deficiency sets, job suitability and their confidence intervals; construct non-compliance scores by mapping the observed capability indicators obtained from subsequent work closed-loop quality, safety violation rate, defect recurrence rate or test retest scores with the prediction residuals between them and the prediction results; take the quantile values ​​of the non-compliance score distribution under the target coverage, and generate the confidence intervals of capability scores and job suitability from the quantile values. S5. Input the set of skill gaps, job fit, confidence interval, training resources, certification period and risk level into the multi-objective programming model. Output the career development path when the job fit reaches the job fit threshold, the confidence interval width does not exceed the confidence interval threshold and the risk level is not higher than the job safety threshold. Otherwise, output the supplementary certification instruction and / or training instruction and / or temporary recommendation instruction.

2. The method for assessing the competence and career development planning of employees in a power grid enterprise according to claim 1, characterized in that, S1 includes: For each piece of employee competency evidence data, an evidence record is generated. The evidence record includes the data source identifier, generation time, update time, business object identifier, event result, and risk level. The same employee, position, task, equipment, and competency tag are mapped to a unified identifier according to a preset field dictionary. Employee, position, task, equipment, risk scenario, competency tag, course, certificate, and business event are used as nodes, and the following are used as edges: appointment, participation, associated equipment, corresponding risk, proven competency, required competency, completed course, and held certificate. The generation time is used as the edge to write a timestamp, resulting in a time-series heterogeneous competency knowledge graph.

3. The method for assessing the competence and career development planning of employees in a power grid enterprise according to claim 1, characterized in that, S4 includes: performing dimensional completion and normalization on the scenario-based employee capability vector according to the dimensional order of the job capability benchmark vector, and calculating the dimensional difference vector; inputting the dimensional difference vector into the scoring mapping table to obtain the capability score and capability weakness set of each capability tag; and inputting the dimensional difference vector into the fit mapping model to obtain the job fit.

4. The method for assessing the competence and career development planning of employees in a power grid enterprise according to claim 1, characterized in that, The heterogeneous graph Transformer is obtained through self-supervised training, and the training samples are generated from the historical time-series heterogeneous capability knowledge graph. The training tasks include occlusion node type recovery, occlusion relationship type recovery, future business event result prediction, and capability label consistency judgment. The training loss is obtained by weighted summation of the loss of each training task and the evidence weight bias regularization term. After training, the model parameters are fixed, and the trained model is used for the vector output in step S3.

5. The method for assessing the competence and career development planning of employees in a power grid enterprise according to claim 3, characterized in that, S5 includes: establishing reachability relationships between capability tags in the capability gap set and courses, certificates, job tasks, and certification cycles to generate candidate development actions; calculating capability enhancement amount, training duration, certification time, and risk constraint satisfaction for each candidate development action; and selecting a sequence of candidate development actions based on the weighted objectives of capability enhancement amount, training duration, and certification time to obtain a career development path when the job fit reaches the job fit threshold, the confidence interval width does not exceed the confidence interval threshold, and the risk level is not higher than the job safety threshold.

6. The method for assessing the competence and career development planning of employees in a power grid enterprise according to claim 5, characterized in that, The supplementary certification instruction, training instruction, or temporary non-recommendation instruction are generated according to objective conditions and the priority of the main instruction: when the risk level is higher than the job safety threshold or the job suitability is lower than the job suitability threshold, the temporary non-recommendation condition is met; when the confidence interval width exceeds the confidence interval threshold and the number of missing evidence reaches the supplementary certification quantity threshold, the supplementary certification condition is met. If the width of the confidence interval does not exceed the confidence interval threshold and the set of capability shortcomings is not empty, the training conditions are deemed met. If the condition of temporary non-recommendation is met, a temporary non-recommendation instruction is generated as the main instruction, and the unmet risk level, job task, and capability tag are output. If both the supplementary certification condition and / or training condition are met, the corresponding supplementary certification instruction and / or training instruction are generated as subordinate instructions and written into the same instruction queue. If the condition of temporary non-recommendation is not met but the condition of supplementary certification is met, a supplementary certification instruction is generated as the main instruction, and the data source identifier and capability tag to be supplemented are output. If both the temporary non-recommendation condition and the supplementary certification condition are not met but the training condition is met, a training instruction is generated as the main instruction, and the course identifier, certificate identifier, and completion deadline are output.

7. A system for assessing the competence and career development of employees in a power grid enterprise, used to execute the method for assessing the competence and career development of employees in a power grid enterprise as described in any one of claims 1 to 6, characterized in that, include: The graph construction module is used to collect employee capability evidence data and generate a time-series heterogeneous capability knowledge graph. The subgraph weight generation module is used to generate scenario query conditions, extract conditional subgraphs, and generate dynamic evidence weights. The vector representation module is used to call the heterogeneous graph Transformer to generate scenario-based employee capability vectors and job capability benchmark vectors; the calibration and evaluation module is used to generate capability scores, capability deficiency sets, job suitability, and confidence intervals. The career planning module is used to generate career development paths, certification requests, training instructions, or instructions not to recommend certain skills.

Citation Information

Patent Citations

  • Tobacco enterprise human resource management auxiliary calibration method based on big data

    CN120689018A