Virtual reality-based human resources remote training system
By collecting and processing multimodal data using virtual reality technology, dynamic twin capability profiles of employees are generated, providing adaptive growth paths. This solves the problems of interactivity and personalization in traditional human resource remote training, and improves training effectiveness and efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHAANXI YANCHANG PETROLEUM MINING CO LTD
- Filing Date
- 2025-12-30
- Publication Date
- 2026-05-29
AI Technical Summary
Traditional remote human resource training lacks interactivity and cannot deliver personalized content based on individual differences among trainees, resulting in low learning efficiency and difficulty in guaranteeing training quality.
A virtual reality-based remote training system for human resources is adopted. Multimodal data is collected through VR interactive devices, cross-modal correlation feature extraction and processing are performed, dynamic twin capability profiles of employees are generated, and personalized training plans are made using an adaptive growth path module, which can be adjusted and optimized in real time.
It enables flexibility and convenience in employee training, allows for dynamic assessment of trainees' knowledge gaps and skill deficiencies, forms adaptive learning paths, and improves learning efficiency and quality.
Smart Images
Figure CN122114871A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of human resource remote training technology, and specifically provides a human resource remote training system based on virtual reality. Background Technology
[0002] Traditional in-person training is costly, and in-person video training lacks context and cannot assess practical skills. Furthermore, traditional training uses standardized content that cannot be adapted to the learning pace and skill level of different employees. Traditional remote HR training also cannot address issues in real time, impacting the quality and efficiency of the training.
[0003] Using standardized training courses, which cannot be personalized based on individual learners' knowledge base, skill levels, and learning progress, leads to low learning efficiency. Both high-level and weak learners struggle to achieve the best learning experience. Essentially, it's a one-way information transmission; remote training lacks real-time interaction, resulting in low training quality. In contrast, virtual reality-based remote human resource training can simulate real-world work scenarios, leading to higher training effectiveness and quality. The system supports training anytime, anywhere, dynamically updates employees' dynamic twin capability profiles, accurately assesses each learner's knowledge gaps and skill deficiencies, and creates adaptive learning paths. Summary of the Invention
[0004] This application provides a virtual reality-based remote training system for human resources to address the problems of lack of interactivity and low efficiency and quality in remote human resources training.
[0005] This application provides a virtual reality-based remote human resources training system, the method of which includes: The multimodal data acquisition and processing module is used to acquire multimodal VR interaction data of users through VR interaction devices. The multimodal VR interaction data includes spatial action sequences, eye tracking data, voice data, and physiological data; and to acquire employee historical training data and target job profiles. The multimodal VR interaction data is subjected to cross-modal correlation feature extraction, which is transformed into structured behavioral data. The structured behavioral data is then cleaned, denoised, time-aligned, and standardized to obtain standardized behavioral data. The standardized behavioral data is then transmitted to the data analysis and evaluation module. The data analysis and evaluation module is used to receive the standardized behavioral data, perform multi-dimensional dynamic evaluation and analysis on the standardized behavioral data, generate scores and evaluation reports for each dimension, and obtain a dynamic twin capability profile of the employee based on the evaluation report and the employee's historical training data. Predictive Analysis and Adaptive Growth Path Module: Input the employee's dynamic twin capability profile and the target job profile, wherein the target job profile includes at least skill requirements and promotion standards, and use a two-layer prediction algorithm to generate an adaptive growth path; The training effectiveness visualization and skills enhancement module displays the evaluation report and the adaptive growth path through a three-dimensional dynamic visualization interface, identifies skills weaknesses, and automatically matches and calls up corresponding specialized training based on the skills weaknesses. The adaptive growth path is broken down into phased training plans, which are then executed by the system. New behavioral data generated during the training process is collected in real time and fed back to the data analysis and evaluation module. The data analysis and evaluation module updates the employee's dynamic twin capability profile based on the new behavioral data and triggers the predictive analysis and adaptive growth path module to dynamically adjust the adaptive growth path.
[0006] In some embodiments, obtaining the employee's dynamic twin capability profile based on the assessment report and the employee's historical training data includes: Based on the assessment report and the employee's historical training data, an initial dynamic capability profile of the employee is generated; Based on the changes in employee capabilities over time, the initial dynamic capability profile of the employee is updated to obtain a dynamic twin capability profile of the employee. Based on the assessment report and the employee's historical training data, an initial dynamic competency profile of the employee is generated. The specific process is as follows: A dual-branch generative adversarial network is used to construct a twin model of job competency benchmarks. The first branch is the real competency mapping branch, which generates an initial twin profile that matches the real competency based on the employee's historical training data and the assessment report as input. The second branch is the benchmark evolution branch, which generates and continuously updates the competency benchmark threshold sequence for the same job group through adversarial learning. Dynamic updates are achieved by using a feature fusion loss function of the first and second branches. Based on the assessment report and the employee's historical training data, a multi-dimensional capability feature vector is extracted, including hard skills, soft skills, and potential and status. Using the multi-dimensional capability feature vectors, an initial dynamic capability profile of the employee is generated through the model.
[0007] In some embodiments, the generation of an adaptive growth path using a two-layer prediction algorithm includes: The two-layer prediction algorithm consists of a gap analysis and growth trend prediction layer and a reinforcement learning dynamic optimization layer. The gap analysis and growth trend prediction layer includes: The target job profile also includes real-time skill requirements and standards for mastering new tools; The gap analysis method is used to compare and calculate the multi-dimensional capability gap values between the employee's dynamic twin capability profile and the target job profile; using a time series prediction model, based on the employee's historical training data and the assessment report, the natural growth trend of the employee in each capability dimension is predicted.
[0008] The virtual reality-based remote training system for human resources provided in this application allows employees to access the training system anytime, anywhere without having to gather at a specific location. This improves the flexibility and convenience of training organization, dynamically updates the dynamic twin capability profiles of employees, accurately assesses each trainee's knowledge gaps and skill deficiencies, and forms an adaptive learning path, thereby improving the learning efficiency, quality, and training effectiveness of each employee. Attached Figure Description
[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0010] Figure 1 This is a flowchart illustrating the virtual reality-based remote training system for human resources provided by the present invention. Detailed Implementation
[0011] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0012] The following is combined with Figure 1 The illustrated embodiments describe the technical solution of the present invention: This application provides an embodiment of a virtual reality-based remote training system for human resources, referring to... Figure 1 As shown, the virtual reality-based remote human resource training system provided in this embodiment includes the following steps: S110: Multimodal data acquisition and processing module, used to acquire multimodal VR interaction data of users through VR interaction devices, extract cross-modal correlation features from multimodal VR interaction data, convert it into structured behavioral data, process the structured behavioral data to obtain standardized behavioral data, and transmit the standardized behavioral data to the data analysis and evaluation module.
[0013] In some embodiments, the implementation of the above subsystem S110 (multimodal data acquisition and processing module, used to acquire multimodal VR interaction data of users through VR interaction devices, extract cross-modal correlation features from the multimodal VR interaction data, convert it into structured behavioral data, process the structured behavioral data to obtain standardized behavioral data; and transmit the standardized behavioral data to the data analysis and evaluation module) may include: It should be noted that, firstly, user and content management is required. The training system administrator is responsible for managing employee basic information, permission allocation, uploading, classifying, and distributing VR courses; and rationally arranging training plans, including the personnel, time, and courses to be trained. Raw data is captured using VR devices, including controller position, eye tracking, and voice, and preprocessed by cleaning, denoising, and timestamp synchronization.
[0014] Build a cloud-native management backend that supports hierarchical permission management for multiple enterprises and departments, and connects to the enterprise human resources system API to automatically synchronize employee job information and match corresponding training courses.
[0015] It should be noted that employees participating in remote human resources training first log in to the VR device system, enter the simulated training scenario, and then engage in immersive learning. Trainees perform practical simulations in the virtual environment, such as customer communication, equipment operation, and computer skills; the training system provides operational guidance to complete individual learning tasks and provides feedback.
[0016] It should be noted that administrators can view various types of data, such as personnel training progress and job skills data, through the system backend. Based on these data, the system generates training reports (individual reports and team or overall analysis reports), including indicators such as completion rate and practical operation accuracy. The system then performs data analysis on these various data to evaluate training effectiveness and update and optimize VR training content. Finally, the employee training data and evaluation results are incorporated into the human resources management system to provide data support for decisions regarding employee career development paths, promotions, and job transfers.
[0017] It should be noted that the multimodal data acquisition and processing module is used to acquire the user's multimodal VR interaction data through VR interaction devices and capture the person's movements in real time. The multimodal VR interaction data includes spatial movement sequences, eye tracking data, voice data, and physiological data. By using built-in sensors and external APIs in VR devices, multimodal interaction data of users in training scenarios can be collected, including spatial motion sequences, eye-tracking data, voice data, and physiological data.
[0018] The collected multi-source data must first have a unified timestamp, and the controller position data, eye-tracking data, and voice stream must be correlated and synchronized.
[0019] It should be noted that VR devices are used to collect VR interaction data from users; employee behavior data (motion capture, reaction time) and skills assessment data generated during training are combined with VR virtual training scenarios and embedded into training course content and interaction logic.
[0020] It's important to note that transforming multimodal data into quantifiable feature indicators is crucial for further data analysis and evaluation. This data directly supports the construction of dynamic twin capability profiles and training assessments for employees. Algorithms transform raw data into behavioral feature data; for example, controller movement trajectories are converted into operational path efficiency, eye-tracking data into attention focus distribution, and voice data is transformed into emotional positivity through STT (Static Transmission Theory Test). Physiological indicators, such as eye-tracking trajectories reflecting focus, behavioral efficiency, and communication skills, are also included.
[0021] Structured behavioral data includes at least hard skill-related structured behavioral data, soft skill-related structured behavioral data, and potential and state-related structured behavioral data. The extraction and transformation process is implemented through a cross-modal temporal alignment algorithm (timestamp error ≤ 10ms), and the specific process is as follows: Structured behavioral data related to hard skills: 1) Operational accuracy measurement: The accuracy is generated by calculating the average Euclidean distance (deviation) between the three-dimensional coordinates (x, y, z) of the spatial action sequence and the standard operation trajectory in the virtual scene, according to the formula: accuracy = 1 - (average distance / maximum allowable deviation) (the maximum allowable deviation is preset according to the operation type, and the fine operation is ≤3cm). For example, 1) the collected virtual tool uses a trajectory (three-dimensional coordinate sequence) and is timestamped with the standard trajectory (to ensure coordinate comparison at the same moment); 2) the Euclidean distance deviation at each time point is calculated; 3) the average deviation rate of the whole process is calculated; 4) based on the deviation rate distribution of the employee's historical training data (e.g., the 90th percentile is 5%), a score of 90 is set for deviation rate ≤ 5%, 70 for 5%-10%, and 50 for deviation rate > 10%. 2) Process compliance measurement: Corresponding to the coherence of spatial action sequences and the accuracy of voice commands, the percentage of operation steps completed according to the standard process is statistically analyzed to obtain the probability of task process compliance. For example, a standard task flow is preset (e.g., 10 mandatory steps, in the order S1→S2→…→S10, each step corresponding to a standard voice command "command S1", "command S2", etc.); the actual operation step sequence (S'1, S'2,..., S'm) and voice command sequence (I'1, I'2,..., I'm) are recorded in real time; compliance is determined by using cosine similarity to calculate the matching degree between the step sequence S1→S2→…→S10 and the voice command I'k and the standard command of the corresponding step Sk. If the matching degree is ≥90%, then it is compliant.
[0022] 3) Information Processing Efficiency: Efficiency = Information Focusing Duration of Eye-Tracking Data / Complexity Coefficient of Virtual Scene Information Elements (The complexity coefficient is a weighted value of information hierarchy, data volume, and interaction correlation, ranging from 1.0 to 5.0). By linking the information fixation duration, information extraction efficiency, and spatial motion response speed in the eye-tracking data, the time from information extraction to the execution of the corresponding operation is obtained. The average time for each training session is calculated to obtain the information processing efficiency. Preset efficiency thresholds based on job requirements, such as expert calibration: Level A ≤ 0.8 seconds, Level B 0.8-1.2 seconds, Level C > 1.2 seconds. For example, a fixation time of 0.7 seconds for a correct operation results in a Level A information processing efficiency.
[0023] Structured behavioral data related to soft skills: 1) Collaborative response efficiency: The command recognition result of the voice data is aligned with the corresponding spatial action timestamp. Efficiency = 1 / (command end timestamp - action start timestamp) × collaborative scenario difficulty coefficient (difficulty coefficient: low = 1.0, high = 2.0). 2) Emotional resilience score and emergency response rationality: By constructing a collaborative change model between speech acoustic features (speech rate, pitch fluctuation, volume change rate) and physiological data such as heart rate variability (HRV) and skin conductance signal amplitude during stress events (the correlation is calculated using the Pearson correlation coefficient, and the time series correlation coefficient and dynamic response lag time are calculated), emotional resilience = 0.6 × (1 - HRV fluctuation variance / baseline value) + 0.4 × (1 - skin conductance signal fluctuation amplitude / threshold), the emotional resilience score (based on the synergy between physiological signal fluctuation amplitude and speech stability) and emergency response rationality (based on the matching degree between speech command logic and physiological stress intensity) are output. The emergency response rationality is generated based on the cosine similarity between the above-mentioned correlation features and the preset optimal emergency mode.
[0024] By linking eye-tracking data with information fixation duration, information extraction efficiency, and spatial motion response speed, the time from information extraction to execution of the corresponding operation is obtained. The average time for each training session is calculated to obtain the information processing efficiency. Efficiency thresholds based on job requirements are preset, such as expert calibration: A-level ≤ 0.8 seconds, B-level 0.8-1.2 seconds, and C-level > 1.2 seconds. For example, if the correct operation is executed after a fixation of 0.7 seconds, the information processing efficiency is A-level.
[0025] It should be noted that the stress resistance score is generated based on physiological data (heart rate, respiration) and fluctuations in tone and speech rate in voice data. Specifically, a baseline value is collected in a calm state, then the fluctuation value in a stressed state is calculated, and finally a weighted sum is obtained. The weights are designed based on expert experience, and the higher the score, the stronger the resilience.
[0026] By correlating the skin conductance response intensity, heart rate variability, and movement adjustment direction in spatial movement sequences with physiological data, we can determine the similarity between the action processing and the optimal solution in sudden situations such as task adjustment or equipment failure, and obtain the rationality of emergency response. For example, if the similarity is 78%, the rationality is rated as moderate.
[0027] For example, physiological data (skin conductance response intensity GSR, heart rate variability HRV), spatial movement sequence, and preset optimal emergency response trajectory A_opt; When the skin conductance response intensity (GSR) exceeds twice the calm value and the heart rate response (HRV) decreases by 30%, an emergency state is triggered. The action adjustment trajectory A_act under the emergency state is obtained, and the time series of A_act and A_opt are aligned using the Dynamic Time Warping (DTW) algorithm to obtain the similarity Sim = (1 - DTW distance / maximum possible distance) × 100%. Sim ≥ 80% is excellent, 60%-80% is moderate, and < 60% is in need of improvement.
[0028] It should be noted that by analyzing eye-tracking data, physiological data, and historical operation sequences across modal time series, attention concentration is generated based on the target area focusing ratio of eye-tracking data; stability under pressure is generated based on the fluctuation variance of physiological data within the high-pressure task time interval in the virtual scene; and learning ability is generated based on the temporal improvement rate and error rate decline trend of task completion efficiency in historical operation sequences.
[0029] Structured behavioral data on potential and state: 1) Attention concentration: Concentration = Target area focus time of eye tracking data / Total interaction time × Importance weight of target area (Importance weight: Core area = 0.8, Auxiliary area = 0.2); 2) Stability under pressure: Based on physiological data (heart rate, respiratory rate) in a virtual high-pressure task (pressure gradient ≥80%), stability = 1 - (actual variance / preset pressure baseline variance). 3) Learning ability: The task completion efficiency improvement rate (slope k1) and error rate decrease trend (slope k2) are fitted by linear regression of historical operation sequences. Learning ability = 0.5×k1+0.5×(1-|k2|), where k1 and k2 are calculated based on the data of the most recent 30 operations.
[0030] Specifically, the slope of the learning curve is calculated by relating the operation duration and the number of errors (spatial action errors and speech errors) of multiple repetitive tasks in the spatial action time sequence to determine the rate of change of ability per unit time.
[0031] For example, the spatial action sequence (operation duration T_k of repeated tasks), the number of errors (spatial action error E_act_k + voice error E_voice_k, total error E_k=E_act_k+E_voice_k), k=1,2,...,n (training times); Specifically: ① With training times k as the horizontal axis and total error rate R_k=E_k / total number of steps×100% as the vertical axis, fit the linear regression equation: R_k=a×k+b; ② Learning curve slope=-a (a positive slope indicates a decrease in error rate, i.e., an improvement in ability. The larger the slope, the faster the improvement. For example, after 3 training sessions, the error rate drops from 50% to 20%, and the slope is (50-20) / 3=10% / time).
[0032] Cross-modal correlation features are extracted from multimodal VR interaction data and transformed into structured behavioral data. The structured behavioral data is then cleaned, denoised, time-aligned, and standardized to obtain standardized behavioral data. The standardized behavioral data is then transmitted to the data analysis and evaluation module.
[0033] It should be noted that a distributed cleaning algorithm based on federated learning is used to collaboratively construct a global noise model without centralizing the original user data, in order to eliminate the acquisition bias introduced by the differences between various VR terminal devices.
[0034] Specifically, under the federated learning framework, structured behavioral data distributed across various user terminals undergoes localized cleaning. The cleaning process includes removing outliers exceeding preset thresholds (such as abnormal physical boundaries of spatial action coordinates or abnormal numerical ranges of physiological signals), filling missing values based on interpolation of adjacent time-series data, and deleting duplicate records. The cleaned structured behavioral data is then subjected to multimodal consistency verification (verification dimensions include: whether the timestamp synchronization deviation of different modal data is within a preset range (e.g., ≤100ms), whether the spatial action data matches the interaction logic of the virtual scene, and whether the semantic matching degree between voice commands and the corresponding spatial actions reaches a preset threshold (e.g., ≥80%). Based on the verification results, the data undergoes time-series alignment processing (unifying the multimodal data to the same time axis, with a time granularity accurate to a preset millisecond level of 50ms), and data with excessive time deviations are linearly corrected. The time-aligned structured behavioral data is denoised using a sliding window filtering method to remove high-frequency noise. The sliding window size is dynamically set based on the task interaction frequency; for example, the window for high-frequency tasks is 5 time units, and the window for low-frequency tasks is 10 time units. The denoised structured behavioral data is then dynamically standardized: time windows are divided based on task scenario types, such as high-pressure tasks and collaborative tasks. Within each window, Z-score standardization (based on the mean and standard deviation of the behavioral data of the user group within that window) is applied. Spatiotemporal weight coefficients are introduced during the standardization process. The time weight is a decay factor based on the data generation time and the current analysis time, with the decay coefficient set based on the task's timeliness. The spatial weight is the importance coefficient of the interaction area in the virtual scene (based on the correlation between the area and the task objective). The standardization results are multiplicatively weighted and corrected using the spatiotemporal weight coefficients to generate standardized behavioral data with spatiotemporal weights. This standardized behavioral data is then transmitted to the data analysis and evaluation module.
[0035] It should be noted that the dynamic standardization algorithm is used, and its normalization parameters are dynamically adjusted based on the difficulty of the current training scenario and the user's historical ability level to ensure that the ability scores under different difficulty scenarios are comparable.
[0036] It should be noted that spatial motion sequences include limb movement precision, operation timing continuity, and virtual tool usage trajectory; eye-tracking data includes key information fixation duration, attention shift path, and virtual scene information capture efficiency; voice data includes the accuracy of collaborative instructions, communication response latency, and information transmission integrity; and physiological data includes heart rate variability, skin conductance intensity, and respiratory rate variability.
[0037] For example, the timing of spatial movement errors is time-series aligned with peak physiological stress and eye fixation point shift to generate associated behavioral labels of operational errors, emotional fluctuations, and attention distraction, which are then transformed into structured behavioral data. After preprocessing the multimodal data, drift errors from VR devices are removed through dynamic noise filtering. Multimodal time-series alignment (unifying the timeline of data acquisition from different devices) and standardization are then performed, and the processed multimodal data is transmitted to the data analysis and evaluation module.
[0038] It should be noted that it is also necessary to ensure that network resources are allocated reasonably and training content is comprehensive, that all personnel see the same status in the same virtual space, and that multi-person remote interaction can be used for group discussions to train employees' communication skills and problem-solving abilities.
[0039] It should be noted that virtual reality remote human resource training can simulate real-world scenarios, making the training content more realistic and interactive, thereby improving the quality of training.
[0040] S120: Data Analysis and Evaluation Module, used to receive standardized behavioral data, perform multi-dimensional dynamic evaluation and analysis on the standardized behavioral data, and generate evaluation reports; based on the evaluation reports and employees' historical training data, to obtain employees' dynamic twin capability profiles.
[0041] In some embodiments, the implementation of the above subsystem S120 (data analysis and evaluation module, used to receive standardized behavioral data, perform multi-dimensional dynamic evaluation and analysis on the standardized behavioral data, generate an evaluation report; and obtain a dynamic twin capability profile of employees based on the evaluation report and employee historical training data) may include: It should be noted that algorithms are used to analyze and evaluate employee behavioral data, outputting quantifiable scores and generating assessment reports. VR interactive data is transformed into quantifiable skill scores, allowing for an objective assessment of training effectiveness.
[0042] It should be noted that quantitative indicators are generated by quantifying the feature values of multi-dimensional related structured behavioral data.
[0043] Hard skills assessment: 1) Virtual task completion rate: Based on the ratio of the number of actually completed task nodes to the total number of nodes in the preset task process (number of completed nodes / total number of nodes × 100%). 2) Operation standardization rate: The percentage of matching steps between spatial action sequences and standard operation templates in virtual scenes (number of matching steps / total number of steps × 100%). 3) Error correction efficiency: Analyze the time interval between the occurrence of an operational error and the execution of a corrective action to evaluate the error correction efficiency. The efficiency is based on the time interval between the occurrence of the operational error and the completion of the corrective action (the smaller the interval, the higher the efficiency, quantified as 0-100 points). Soft skills dimension assessment: 1) Collaborative response sensitivity: Based on the reciprocal of the time difference between the moment of receiving a voice command and the moment of initiating the corresponding action in a multi-person collaborative scenario (the smaller the time difference, the higher the sensitivity, standardized to 0-100 points). 2) Rationality of emergency decision-making: The semantic sentiment of the speech data (identified by the BERT model) and the heart rate variability (HRV) of the physiological data are fused to generate an emotion-physiological stress correlation feature vector; a score of 0-100 is generated by weighting the matching degree between the operation sequence in the stress event and the optimal decision path (accounting for 60%) and the speech semantic coherence (calculated by the BERT model with the cosine similarity to standard semantics, accounting for 40%). 3) Context adaptability: Based on the cumulative attenuation coefficient of operational deviation (current deviation value / historical average deviation value, coefficient <1 indicates improved adaptability).
[0044] Potential and Status Dimensions Assessment: 1) Learning curve slope for new scenarios: The slope is calculated using linear regression with task completion time as the dependent variable and training times as the independent variable (the slope is negative and the larger the absolute value, the higher the learning efficiency). 2) Behavioral stability under stress: The growth rate of the standard deviation of physiological data (heart rate, skin conductance) fluctuations in a virtual scene stress gradient (divided into 5 levels according to complexity) (growth rate <10% is considered stable, quantified as 0-100 points). 3) Error pattern recognition speed: based on the reciprocal of the time interval between the first and second occurrences of the same type of error (the shorter the interval, the faster the recognition speed, standardized to 0-100 points).
[0045] It should be noted that, based on the historical time series of standardized behavioral data, the learning curve is dynamically fitted (the slope of the curve is updated in real time to reflect changes in short-term learning efficiency), and combined with the variance of physiological data fluctuations and the recurrence interval of error patterns under the pressure gradient of the virtual scene (gradually increasing from low to high), the potential dimension score and the prediction of learning ability growth are dynamically output.
[0046] It should be noted that the evaluation report generates cross-dimensional assessment results and targeted improvement suggestions through dynamic correlation rules of the three dimensions of scores (such as simultaneously reducing the weight of the soft skills emergency decision-making rationality score when the hard skills error correction efficiency is lower than the threshold).
[0047] The assessment report should include at least a dynamic skills composite score and a cross-dimensional collaborative assessment, in which: Dynamic Skills Overall Score: Based on the pre-set weights of each dimension according to the training objectives, the scores of each indicator in the dimensions of hard skills, soft skills, potential and status are weighted and summed.
[0048] For example, based on the preset weights of the training objectives, such as 60% for hard skills, 20% for soft skills, and 20% for potential and status in technical training, the scores of each dimension (0-100 points) are weighted and summed: Total score = Hard skills score × 60% (weight 1) + Soft skills score × 20% (weight 2) + Potential and status score × 20% (weight 3). Cross-dimensional linkage assessment: Based on preset dynamic association rules, when any dimension indicator is lower than the preset threshold of 60 points, the system automatically adjusts the weight of its associated dimensions and recalculates the dynamic skill comprehensive score to more comprehensively reflect the trainee's true ability status (e.g., when hard skill error correction efficiency is <60 points), the weight of associated dimensions is adjusted as follows: the weight of soft skill emergency decision-making rationality is reduced by 20%, and the weight of potential and state error pattern recognition speed is increased by 10%, and the basis and impact of the adjustment are noted in the assessment report.
[0049] It should be noted that the difference between the operator's operation and the standard operation is judged. The smaller the difference, the higher the skill score. Based on the changes of each employee, an individual ability profile is generated and updated. The individual ability profile contains a data model of the scores of each skill.
[0050] It should be noted that by integrating historical training data and real-time behavior in VR scenarios, a dynamic twin capability profile of employees is constructed; the capability profile of an employee can be updated in real time with the recommendation of virtual tasks, and the capability status at any point in time can be traced.
[0051] Based on the assessment report and historical training data, generate an initial dynamic capability profile of the employee; Based on the changes in employee capabilities over time, the initial dynamic capability profile of employees is updated to obtain a dynamic twin capability profile of employees. Specifically, based on the assessment report and historical training data, an initial dynamic competency profile of the employee is generated. The specific process is as follows: A dual-branch generative adversarial network (DAN) is used to construct a twin model of job competency benchmarks. The first branch is the real competency mapping branch, which uses employees' historical training data and assessment reports as input to generate an initial twin profile that matches the real competency. The second branch is the benchmark evolution branch, which generates and continuously updates the competency benchmark threshold sequence for the same job group through adversarial learning. By using the feature fusion loss function of the two branches, the employee twin profile can maintain individual specificity while dynamically aligning with the job competency benchmark, realizing dynamic updates of two-way calibration between individuals and the group.
[0052] Based on the assessment report and historical training data, a multi-dimensional capability feature vector is extracted, which includes hard skills, soft skills, potential and status. By utilizing multi-dimensional capability feature vectors, this model generates an initial dynamic capability profile of employees.
[0053] The Temporal-GAN model is used to capture changes in employee capabilities over time, enabling dynamic updates of the twin profile. The model's loss function incorporates a job capability decay coefficient to adapt to the differences in skill iteration speeds for different job positions.
[0054] Based on the latest training interaction data and evaluation reports of employees, the optimized Temporal-GAN model is used to incrementally update the dynamic capability profile of employees, resulting in a dynamic twin capability profile of employees.
[0055] It should be noted that the Temporal-GAN model is continuously optimized through adversarial training to improve its accuracy in capturing changes in employee capabilities. The final output includes a three-dimensional assessment report that includes skill attainment rate, collaboration shortcomings, and potential tendencies, while simultaneously constructing a dynamically updated digital twin profile of employee capabilities.
[0056] S130: Predictive Analysis and Adaptive Growth Path Module, used to input dynamic twin capability profiles of employees and target job profiles; uses a two-layer prediction algorithm to generate adaptive growth paths.
[0057] In some embodiments, the implementation of the above subsystem S130 (predictive analysis and adaptive growth path module, used to input dynamic twin capability profiles of employees and target job profiles; and to generate adaptive growth paths using a two-layer prediction algorithm) may include: It should be noted that the predictive analysis and adaptive growth path module works as follows: Based on training data and career goals, the prediction module is synchronized with the job profile (skill requirements, promotion standards), and an adaptive growth path is generated using a predictive algorithm. The evaluation results, employee personal ability profiles, and job goal profiles are all input into the predictive analysis and adaptive growth path module to generate a personalized training and growth path. The adaptive growth path is then transformed into a specific training plan, and corresponding training is conducted through the skills enhancement module, with the path being dynamically updated.
[0058] It should be noted that targeted training is conducted based on individual abilities, and dynamic development training paths and plans are customized for each employee to achieve high-quality training results and efficiency.
[0059] The prediction algorithm takes into account the employee's personal competency profile, assessment report, career goals, and target job profile. The target job profile includes at least the skill requirements and promotion standards. The algorithm then uses the prediction algorithm to generate an adaptive growth path that includes a training content list, skill enhancement optimization levels, and a cycle plan.
[0060] Employee basic data includes job information, career goals, personal traits, and behavioral data, with the behavioral data being dynamic.
[0061] Job profile data: Retrieve job competency models built from the human resources system, including a list of essential skills, skill level standards, and promotion requirements; It should be noted that, firstly, decision tree or random forest algorithms are used to convert employees' VR interaction data into quantifiable competency indicators for scoring, and these are compared with job requirements to identify gaps. These gaps are then clustered and categorized to obtain levels. Different levels are then processed accordingly to form adaptive growth paths. High-level employees receive training aimed at promotion, mid-level employees receive skills enhancement, and low-level employees continue learning skills or are eliminated.
[0062] Gap analysis algorithm: Compare an individual's competency profile with the competency requirements of the target position, identify the individual's competency gaps, provide targeted training, and adjust the personalized learning path map.
[0063] It should be noted that, based on the latest assessment results, the gap is recalculated, and subsequent learning paths are adjusted to achieve adaptive learning, thereby improving training quality. The status of training content and changes in learners' abilities are tracked in real time, dynamically optimizing the training content and adjusting learning paths.
[0064] The reinforcement learning algorithm DQN is used to generate an initial growth path, in which the skill enhancement efficiency is set as the reward function and the rate of change of job requirements is set as the dynamic decay factor. A two-layer prediction algorithm with a gap analysis and growth trend prediction layer and a reinforcement learning dynamic optimization layer; The gap analysis and growth trend prediction layer includes: The target job profile also includes real-time skill requirements and standards for mastering new tools; The gap analysis method is used to compare and calculate the multi-dimensional capability gap between the dynamic twin capability profile of employees and the target job profile; the time series prediction model is used to predict the natural growth trend of employees in each capability dimension based on historical training data and assessment reports. Specifically, the skill gap vector between the dynamic twin capability profile of employees and the target job profile is calculated by using an improved cosine similarity algorithm. The hard skill gap weight is calculated as (core skill requirement intensity of the job × 0.7 + current skill attainment rate × 0.3), and the soft skill gap weight is dynamically linked to employee collaboration data (e.g., the soft skill gap weight is reduced by 5% for every 10% increase in the success rate of collaborative tasks).
[0065] The reinforcement learning dynamic optimization layer includes: Using the predicted values of the capability gap and growth trends as inputs, a Markov decision process (MDP) model is constructed, where: Obtain the current skill level vector and skill gap vector of employees, and obtain VR training programs to be selected; State is defined as an employee's current ability level vector and ability gap vector; An action is defined as a VR training item to be selected; The reward function is designed to minimize the composite score of the capability gap vector, with skill enhancement efficiency as a key weighting factor. The Deep Deterministic Policy Gradient (DDPG) algorithm is used to solve the MDP model to generate an adaptive growth path that includes training content, suggested duration, and assessment nodes. During the algorithm training process, the rate of change of job requirements is introduced as a dynamic decay factor to adapt to the differences in the iteration speed of skills for different jobs, making the generated growth path more forward-looking and robust.
[0066] It should be noted that a reinforcement learning algorithm is used, with the gap value as input. Based on the optimal improvement path of similar gaps in historical training data and real-time behavioral feedback, the algorithm predicts the marginal benefit of different VR scenarios (including low / medium / high difficulty levels) for ability improvement and outputs a priority ranking of scenarios.
[0067] It should be noted that minimizing the gap vector is the reward function, and the growth path is decomposed into a Markov decision process of state (current ability) - action (training program) - feedback (skill improvement rate). An adaptive path containing training content - duration - assessment nodes is generated through the Deep Deterministic Policy Gradient (DDPG) algorithm.
[0068] It should be noted that, with the goal of bridging capability gaps, the system simulates the effects of different training paths and generates data including key milestones, recommended training activities, and capability improvement priorities. The system can then be optimized based on employee training data, such as optimizing skill weights for different positions, emphasizing different training content for each position, and developing different training plans.
[0069] S140: Based on the training effect visualization and skills enhancement module, it is used to display the evaluation report and adaptive growth path through a three-dimensional dynamic visualization interface, locate skills weaknesses, and automatically match and call up the corresponding special training based on the skills weaknesses.
[0070] In some embodiments, step S140 (based on the training effect visualization and skills enhancement module, used to display the evaluation report and adaptive growth path through a three-dimensional dynamic visualization interface, and to locate skills weaknesses, and automatically match and call corresponding specialized training according to the skills weaknesses) includes: The training effectiveness visualization and predictive analysis module integrates VR interactive data and assessment results to visually display the skill development curves of personnel. Based on individual abilities and job requirements, it predicts personnel development and job suitability, matching suitable positions and providing data support for human resource decisions. When the match between personnel and positions is low, it's advisable to consider reassigning personnel or making appropriate optimizations, thus rationally allocating human resources and contributing to the company's or organization's development.
[0071] It should be noted that the training effectiveness visualization and skills enhancement module displays the results of the evaluation report through the visualization module. The skills radar chart shows the comparison of comprehensive abilities, and the heat map shows the areas of operational focus. Based on the skills radar chart, skills analysis is carried out, and reinforcement learning is carried out on the weak areas to achieve skills enhancement.
[0072] It should be noted that visual displays allow employees and managers to clearly see training results and areas for improvement, enabling targeted practice of weak skills and enhancing overall employee skills.
[0073] It is necessary to automatically record data such as employees' operation status, completion time, number of errors, and skill scores. This allows employees to understand their weaknesses and provides companies with data on employees' skill mastery and overall capabilities, helping the human resources department to assess the effectiveness of training.
[0074] It should be noted that, based on employees' historical training progress and fundamental skill weaknesses, a personalized training sequence is generated through a path planning algorithm; according to the training sequence, the corresponding VR-specific training scenarios are sequentially called and loaded into the virtual reality interaction module.
[0075] The multidimensional assessment data in the assessment report is analyzed hierarchically using an improved decision tree model to pinpoint skill weaknesses; error patterns in the operation sequence and deviation data from the standard process are extracted; and error patterns and deviation data are matched with a pre-built skill deficiency knowledge base to identify at least one target skill weakness.
[0076] Specifically, a 3D capability-path fusion visualization model is constructed: The scores for each dimension are mapped to three-dimensional spatial coordinates, with the origin of the coordinate system serving as the baseline threshold for job competence. Each dimension's score corresponds to a dynamic node on the coordinate axis. Meanwhile, the adaptive growth path is broken down into segmented dynamic trajectories, which include trajectory slope, color matching training difficulty gradient, and improvement priority. The 3D visualization interface supports multi-dimensional interactive linkage. The 3D visualization interface is linked in real time with the employee's dynamic twin capability profile and adaptive growth path. When an employee completes specialized training or generates a new assessment report, the coordinates of the dynamic nodes are synchronously iterated in real time.
[0077] For example, the visualization interface supports multi-dimensional interactive linkage, triggering operations through VR device gestures or terminal touch: clicking on a capability node displays detailed assessment data and entry points for specialized training; dragging the growth trajectory zooms in and out of the timeline and visualizes training progress and capability curve changes; the interface is linked in real time with the employee's dynamic twin capability profile and adaptive growth path, and the 3D model is synchronously iterated when training data is updated, highlighting dimensions of significant capability improvement or high-priority nodes to be completed; it provides multiple perspectives such as global overview, local focus, and group comparison, adapting to the differentiated needs of training managers for overall analysis and employees for self-improvement.
[0078] It should be noted that the assessment report is presented through a skills radar chart and a weakness trend curve. The skills radar chart is used to locate skills weaknesses and match them with corresponding VR-specific training scenarios, such as equipment operation error correction scenarios and collaborative division of labor simulation scenarios. The adaptive growth path is broken down into phased training plans, and training is carried out through the skills enhancement module. Training data is collected in real time and fed back to the data analysis and assessment module, and the growth path is dynamically updated.
[0079] For example, a 3D skill radar chart is displayed on the user interface, allowing employees to view the impact of their current actions on their skill scores in real time. A dynamic heatmap shows the distribution of team skills, and a pre-built VR training scenario-skill element correlation graph is constructed (e.g., hard skills for equipment failure emergency scenarios: operational proficiency + soft skills: emergency decision-making). When a skill weakness is identified, a graph neural network (GNN) is used to mine scenario nodes (correlation degree ≥ 0.8) that are strongly correlated with the weakness in the graph, and scenarios containing multimodal interaction requirements are prioritized (e.g., composite scenarios that require voice commands, fine motor skills, and stable physiological states). During specialized training, error pattern sequences of employees in the scenarios are collected in real time, and the error recurrence probability is predicted using an LSTM model. When the probability is > 60%, the training time for that scenario is automatically extended.
[0080] The adaptive growth path is broken down into phased training plans, which are then executed by the system. New behavioral data generated during the training process is collected in real time and fed back to the data analysis and evaluation module. The data analysis and evaluation module updates the employee's dynamic twin capability profile based on the new behavioral data and triggers the predictive analysis and adaptive growth path module to dynamically adjust the adaptive growth path.
[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A virtual reality based human resource remote training system, characterized in that, include: The multimodal data acquisition and processing module is used to acquire the user's multimodal VR interaction data through the VR interaction device. The multimodal VR interaction data includes spatial motion sequences, eye tracking data, voice data, and physiological data. Obtain employee historical training data and target job profiles; The multimodal VR interaction data is subjected to cross-modal correlation feature extraction, which is transformed into structured behavioral data. The structured behavioral data is then cleaned, denoised, time-aligned, and standardized to obtain standardized behavioral data. The standardized behavioral data is then transmitted to the data analysis and evaluation module. The data analysis and evaluation module is used to receive the standardized behavioral data, perform multi-dimensional dynamic evaluation and analysis on the standardized behavioral data, generate scores and evaluation reports for each dimension, and obtain a dynamic twin capability profile of the employee based on the evaluation report and the employee's historical training data. Predictive Analysis and Adaptive Growth Path Module: Input the employee's dynamic twin capability profile and the target job profile, wherein the target job profile includes at least skill requirements and promotion standards, and use a two-layer prediction algorithm to generate an adaptive growth path; The training effectiveness visualization and skills enhancement module displays the evaluation report and the adaptive growth path through a three-dimensional dynamic visualization interface, identifies skills weaknesses, and automatically matches and calls up corresponding specialized training based on the skills weaknesses. The adaptive growth path is broken down into phased training plans, which are then executed by the system. New behavioral data generated during the training process is collected in real time and fed back to the data analysis and evaluation module. The data analysis and evaluation module updates the employee's dynamic twin capability profile based on the new behavioral data and triggers the predictive analysis and adaptive growth path module to dynamically adjust the adaptive growth path.
2. The virtual reality based remote training system for human resources as claimed in claim 1 wherein, The step of extracting cross-modal correlation features from the multimodal VR interaction data and converting it into structured behavioral data includes: The structured behavioral data includes at least hard skill-related structured behavioral data, soft skill-related structured behavioral data, and potential and state-related structured behavioral data. The extraction and transformation process is achieved through cross-modal temporal alignment, specifically as follows: Obtain the standard operation trajectory of the virtual scene, the preset process of the virtual scene, and the complexity coefficient of the information elements in the virtual scene; The structured behavioral data related to hard skills includes: generating an operation accuracy metric based on the deviation between the spatial action sequence and the standard operation trajectory of the virtual scene at the same time point; generating a process compliance metric based on the temporal matching degree between the spatial action sequence and the preset process of the virtual scene; and using the ratio of the information focusing duration of the eye-tracking data to the complexity coefficient of the information elements in the virtual scene as the information processing efficiency. The soft skills-related structured behavioral data: Based on the time difference between the instruction issuance time and the corresponding spatial action execution time of the voice data, collaborative response efficiency is generated; a temporal collaborative change model of the acoustic features of the voice data and the heart rate and skin conductance signals in the physiological data within the stress event time window is constructed to output emotional resilience score and emergency handling rationality. The potential and state structured behavioral data: Through cross-modal temporal correlation analysis of the eye-tracking data, physiological data, and historical operation sequences, attention concentration is generated based on the target area focusing ratio of the eye-tracking data; stability under pressure is generated based on the fluctuation variance of the physiological data within the high-pressure task time interval in the virtual scene; and learning ability is generated based on the temporal improvement rate and error rate decline trend of task completion efficiency in the historical operation sequence.
3. The virtual reality-based remote human resource training system according to claim 1, characterized in that, The process of cleaning, denoising, temporal alignment, and standardizing the structured behavioral data to obtain standardized behavioral data includes: Within the federated learning framework, the structured behavioral data distributed across various user terminals undergoes localized cleaning. The cleaned structured behavioral data is then subjected to multimodal consistency verification, checking timestamp synchronization and cross-modal logical correlation. Based on the consistency verification, the qualified structured behavioral data undergoes time-series alignment. The time-series aligned structured behavioral data is then denoised using sliding window filtering. The denoised structured behavioral data is then dynamically standardized, introducing spatiotemporal weight coefficients during the standardization process. These coefficients are used to perform multiplicative weighted correction on the standardization results, generating standardized behavioral data with spatiotemporal weights. This standardized behavioral data is then transmitted to the data analysis and evaluation module.
4. The virtual reality-based remote human resource training system according to claim 1, characterized in that, The standardized behavioral data is subjected to multi-dimensional dynamic evaluation and analysis to generate scores and evaluation reports for each dimension, including: The standardized behavioral data includes structured behavioral data related to hard skills, structured behavioral data related to soft skills, and structured behavioral data related to potential and status. The standardized behavioral data is processed using a pre-defined multi-dimensional evaluation model and dynamic association rules to output scores for each dimension and an evaluation report, wherein: Hard skills dimension assessment: By processing the feature quantification values of the structured behavioral data related to the hard skills, the virtual task completion rate, operation standardization rate, and error correction efficiency are obtained; Soft skills dimension assessment: By processing the feature quantification values of the structured behavioral data related to the soft skills, we can obtain the sensitivity of collaborative response, the rationality of emergency decision-making, and the adaptability of the situation. Potential and State Dimension Evaluation: By processing the feature quantification values of the structured behavioral data of potential and state, the slope of the new scenario learning curve, behavioral stability under stress, and error pattern recognition speed are obtained; The assessment report shall include at least a dynamic skills comprehensive score and a cross-dimensional collaborative assessment. The dynamic skill comprehensive score is specifically obtained by: acquiring the scores of each dimension of hard skills, soft skills, potential and status; pre-setting the weights of each dimension based on the training objectives; and weighting and summing the scores of each indicator of the hard skills, soft skills and potential dimensions to obtain the dynamic skill score. The cross-dimensional linkage assessment: Based on preset dynamic association rules, when the score of any dimension indicator is lower than the preset threshold, the system adjusts the weight of the associated dimension and regenerates the dynamic skill comprehensive score.
5. The virtual reality-based remote training system for human resources according to claim 1, characterized in that, The process of obtaining a dynamic twin capability profile of employees based on the assessment report and the employees' historical training data includes: Based on the assessment report and the employee's historical training data, an initial dynamic capability profile of the employee is generated; Based on the changes in employee capabilities over time, the initial dynamic capability profile of the employee is updated to obtain a dynamic twin capability profile of the employee. Based on the assessment report and the employee's historical training data, an initial dynamic competency profile of the employee is generated. The specific process is as follows: A dual-branch generative adversarial network is used to construct a twin model of job competency benchmarks. The first branch is the real competency mapping branch, which generates an initial twin profile that matches the real competency based on the employee's historical training data and the assessment report as input. The second branch is the benchmark evolution branch, which generates and continuously updates the competency benchmark threshold sequence for the same job group through adversarial learning. Dynamic updates are achieved by using a feature fusion loss function of the first and second branches. Based on the assessment report and the employee's historical training data, a multi-dimensional capability feature vector is extracted, including hard skills, soft skills, and potential and status. Using the multi-dimensional capability feature vectors, an initial dynamic capability profile of the employee is generated through the model.
6. The virtual reality-based remote training system for human resources according to claim 5, characterized in that, The process of updating the initial dynamic competency profile of an employee based on changes in employee capabilities over time to obtain a dynamic twin competency profile includes: The Temporal-GAN model is used to capture changes in employee capabilities over time. The loss function of the model incorporates a job capability decay coefficient to adapt to the differences in the iteration speed of skills for different jobs. Based on the latest interaction data and the evaluation report, the optimized Temporal-GAN model is used to incrementally update the initial dynamic capability profile of the employee, thereby obtaining the dynamic twin capability profile of the employee.
7. The virtual reality-based remote training system for human resources according to claim 1, characterized in that, The method of generating an adaptive growth path using a two-layer prediction algorithm includes: The two-layer prediction algorithm consists of a gap analysis and growth trend prediction layer and a reinforcement learning dynamic optimization layer. The gap analysis and growth trend prediction layer includes: The target job profile also includes real-time skill requirements and standards for mastering new tools; The gap analysis method is used to compare and calculate the multi-dimensional capability gap between the employee's dynamic twin capability profile and the target job profile; using a time series prediction model, based on the employee's historical training data and the assessment report, the natural growth trend of the employee in each capability dimension is predicted.
8. The virtual reality-based remote training system for human resources according to claim 7, characterized in that, The reinforcement learning dynamic optimization layer includes: Using the predicted results of the capability gap and the growth trend as inputs, a Markov Decision Process (MDP) model is constructed, where: Obtain the current skill level vector and skill gap vector of employees, and obtain VR training programs to be selected; State is defined as the employee's current ability level vector and ability gap vector; An action is defined as the VR training item to be selected; The reward function is designed to minimize the overall score of the capability gap vector, with skill enhancement efficiency as a key weighting factor; The Deep Deterministic Strategy Gradient (DDPG) algorithm is used to solve the MDP model and generate an adaptive growth path, which includes at least training content, suggested duration, and assessment nodes.
9. The virtual reality-based remote training system for human resources according to claim 1, characterized in that, The process of displaying the evaluation report and the adaptive growth path through a three-dimensional dynamic visualization interface includes: Constructing a 3D capability-path fusion visualization model: The scores of each dimension are mapped to three-dimensional spatial coordinates, with the origin of the coordinates as the benchmark threshold for job competence, and the scores of each dimension correspond to dynamic nodes on the coordinate axes. Meanwhile, the adaptive growth path is decomposed into a segmented dynamic trajectory, which includes trajectory slope, color matching training difficulty gradient, and improvement priority. The three-dimensional dynamic visualization interface supports multi-dimensional interactive linkage. The three-dimensional dynamic visualization interface is linked in real time with the employee's dynamic twin capability profile and the adaptive growth path. When the employee completes specialized training or generates a new assessment report, the coordinates of the dynamic nodes are synchronously iterated in real time.
10. The virtual reality-based remote training system for human resources according to claim 1, characterized in that, The system identifies skill weaknesses and automatically matches and invokes corresponding specialized training based on these weaknesses, including: The multidimensional assessment data in the assessment report is analyzed, and an improved decision tree model is used to perform hierarchical analysis of the multidimensional assessment data in order to locate skill weaknesses. The hierarchical parsing specifically includes: Extract error patterns and deviation data of key indicator achievement from standard procedures in the operation sequence; match the error patterns and deviation data with a pre-set skill deficiency knowledge base to identify at least one target skill weakness; Based on the identified target skill weaknesses, the system automatically matches and retrieves the corresponding specialized training content.