A self-help management system for quality of coal mine workers in a distributable arrangement

Abstract

This invention discloses a distributed self-service management system for coal mine employee competence, relating to the field of employee management technology. It includes: a server, and distributed terminal access devices connected to the server via wired or wireless network communication; a data aggregation and alignment module; an individual anomaly screening module; a virtual-real coupling verification module; a multi-dimensional coupling attribution module; a graded intervention and closed-loop verification module; and a model evolution and iteration module. This invention aggregates assessment process data, job operation behavior data, production rhythm data, and team group data through the data aggregation and alignment module, and performs time-series alignment and feature extraction based on a unified time benchmark. Combined with the dynamic baseline construction and three-dimensional anomaly detection of the individual anomaly screening module, and the coupling verification of assessment scenario data and job operation behavior data by the virtual-real coupling verification module, the accuracy and reliability of the assessment results are improved.

Description

Technical Field

[0001] This invention relates to the field of employee management technology, and in particular to a self-service management system for the quality of coal mine employees that can be distributed and deployed. Background Technology

[0002] Coal mine safety production places extremely high demands on the skills and competence of its employees, involving training and assessment in multiple areas such as self-rescue device donning, required knowledge and skills, standardized job operations, and emergency response. With the development of information technology, coal mining enterprises have gradually introduced employee competency management systems. These systems utilize servers and distributed terminal access devices (such as self-service terminals in the mine entrance attendance room, smartphone scanning terminals, and computers in team meeting rooms) to support employees in conducting self-study and assessments during fragmented time. This has alleviated, to some extent, the problems of high investment, large land area, and poor flexibility associated with traditional centralized computer room training models.

[0003] However, existing systems generally adopt a static evaluation and open-loop management technical architecture, which only makes a binary judgment (meeting the standard / not meeting the standard) based on a single assessment score and a fixed passing score. They lack the ability to aggregate and dynamically analyze multi-source heterogeneous data such as assessment process data, job operation behavior data, production rhythm data, and team group data. As a result, the system cannot identify the decline trend of employees' ability level over time, nor can it distinguish whether the abnormality of employees' assessment is caused by knowledge and skills degradation, physical fatigue, terminal environmental interference, or emotional stress. The authenticity and reliability of the assessment results are insufficient.

[0004] Furthermore, the existing system's assessment scenarios are severely disconnected from the actual job performance scenarios, resulting in a separation between the virtual and real environments. The system only records employees' answers in the simulated assessment environment and does not couple and verify them with the actual job performance recorded in existing information systems such as the mine personnel positioning system and equipment monitoring system (e.g., frequency of hazard reporting, trajectory compliance rate, and equipment misoperation records). This leads to situations where test-taking employees who perform well on the test but not in actual work cannot be identified, and employees who perform poorly on the test but are competent in actual work are misjudged as high-risk. Summary of the Invention

[0005] The purpose of this invention is to provide a self-management system for the quality of coal mine employees that can be distributed to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a self-service management system for the quality improvement of coal mine workers that can be distributed, comprising: A server, and a distributed terminal access device connected to the server via a wired or wireless network, the distributed terminal access device being configured to support coal mine employees' self-service access to the server. The data aggregation and alignment module is used to obtain multi-source heterogeneous data of employees from distributed terminal access devices and existing information systems in the mine, and to perform time-series alignment and feature extraction on the multi-source heterogeneous data based on a unified time benchmark to generate a feature dataset. The individual anomaly screening module is used to perform initial screening of individual quality risks based on feature datasets and generate individual anomaly labels. The virtual-real coupling verification module is used to perform virtual-real coupling verification on individual abnormal labels, and generates risk authenticity labels based on the coupling degree between assessment scenario data and job operation behavior data; The multidimensional coupled attribution module is used to perform multidimensional coupled attribution analysis based on risk authenticity labels. The multidimensional coupled attribution analysis includes time-rhythm coupled attribution, work group clustering scanning and organizational attribution, generating a comprehensive risk profile. The graded intervention and closed-loop verification module is used to implement graded interventions based on the comprehensive risk profile and to perform differentiated closed-loop verification of the intervention effect. The verification results are fed back to the individual anomaly screening module and the virtual-real coupling verification module to iteratively optimize their parameters. The model evolution iteration module is used to perform model evolution iteration based on the results of closed-loop verification and the correlation of incident backtracking.

[0007] Preferably, the multi-source heterogeneous data includes assessment process data collected by distributed terminal access devices, job operation behavior data collected by the mine personnel positioning and equipment monitoring system, production rhythm data collected by the mine production scheduling system, and team group data collected by the employee management database. The data aggregation and alignment module uses the end time of the assessment as the anchor point to extract forward time-series features from the assessment process data, extracts first sliding window features with the end time of the assessment as the endpoint from the job operation behavior data, extracts second sliding window features with the end time of the assessment as the center from the production rhythm data, and establishes a rolling monitoring window for the work group data with the end time of the assessment as the center.

[0008] Preferably, the individual anomaly screening module includes: The dynamic baseline construction unit is used to establish a dynamic baseline library layered by job position, length of service, and skill level. New employees use the same job position and length of service group as a substitute baseline during the first month's assessment cycle. After the first month's assessment cycle ends, it switches to an individual dynamic baseline that is updated on a rolling basis based on the individual's data over the past thirty days. The three-dimensional anomaly detection unit is used to perform vertical anomaly detection, horizontal anomaly detection, and behavioral anomaly detection. Vertical anomaly detection is used to determine the degree of deviation of the current assessment score from the individual's dynamic baseline. Horizontal anomaly detection is used to determine the degree of deviation of the current assessment score from the average of the same position and the same length of service group. Behavioral anomaly detection is used to determine whether there are characteristic degradation patterns in the answering process. The tag generation unit is used to generate individual anomaly tags containing anomaly type, anomaly intensity, and anomaly knowledge modules based on the 3D anomaly detection results.

[0009] Preferably, the virtual-real coupling verification module includes: The trigger judgment unit is used to activate the job operation behavior database for coupled verification only when the abnormality intensity of an individual abnormal label reaches moderate or above. The coupling judgment unit is used to perform the following judgments based on the coupling degree between the assessment scenario data and the job operation behavior data: if the assessment degradation signal is strongly coupled with the job operation behavior degradation signal, a real risk label is generated; if the assessment degradation signal is weakly coupled with the job operation behavior normal signal, a state noise label is generated; if the assessment normal signal is reversely coupled with the job operation behavior abnormal signal, a blind zone risk label is generated. The noise source tracing unit is used to retrieve the terminal environment log of this assessment when a status noise label is generated. When the light intensity is greater than the first threshold and it is a touch screen operation, it is determined to be strong light accidental touch noise. When the ambient noise is greater than the second threshold and it is an audio interactive question, it is determined to be environmental interference noise. When the employee's heart rate variation coefficient is lower than the third threshold and the time spent in the well is greater than the fourth threshold, it is determined to be physiological fatigue noise. When none of the above conditions are met, it is determined to be emotional stress noise. The blind spot handling unit is used to semantically match the abnormal characteristics of job operations with the assessment question bank when a blind spot risk label is generated. If there is no corresponding assessment point, it is marked as an assessment blind spot and triggers targeted supplementation of the question bank and special practical assessment.

[0010] Preferably, the multidimensional coupled attribution module includes a time-rhythm coupled attribution unit, which is used to connect with the mine production scheduling system to establish a production event calendar and to construct an individual production load index by combining the individual shift schedule rhythm of employees; When the risk authenticity label is real risk, the time-rhythm coupled attribution unit performs the following judgments: if the individual productivity load index is greater than the load threshold and the current time point is within the production event window, and the employee has experienced an abnormality more than or equal to three times under similar load conditions in the past six months, it is judged as cumulative damage type; if it is less than three times, it is judged as temporary fatigue type; if the individual productivity load index is normal and the current time point is far from the production event window, it is judged as individual degeneration type; if the individual productivity load index is greater than the load threshold and the current time point is far from the production event window, it is judged as latent overwork type.

[0011] Preferably, the multidimensional coupling attribution module further includes a team group attribution unit, which is used to establish a seven-day rolling monitoring window for the team with the end time of the assessment as the center, and calculate the abnormal clustering index. The abnormal clustering index is equal to the product of the number of abnormal people and the Jaccard similarity of the common blind area. When the abnormal clustering index is less than the clustering threshold, it is marked as an isolated event. When the abnormal clustering index is greater than or equal to the clustering threshold, a three-layer tracing mechanism is initiated: training input tracing, which retrieves recent training records for the work group; when the matching degree between the training content and the abnormal knowledge module is lower than the matching threshold, it is determined to be a training defect infection; work group behavior tracing, which analyzes underground collaboration data; when the homogenization rate of hazard photos is higher than the homogenization threshold and there is a pairing trajectory between old and new employees, it is determined to be a negative demonstration infection; and external environment tracing, which retrieves records of mine equipment changes and procedure adjustments; when the abnormal module is strongly correlated with recent changes, it is determined to be an external change infection.

[0012] Preferably, the multidimensional coupled attribution module further includes an attribution fusion decision unit, used to construct an attribution fusion decision matrix, cross-mapping the attribution results of the time-rhythm coupled attribution unit with the attribution results of the work group attribution unit to generate a comprehensive risk level and a precise intervention plan. When the time attribution result is temporary fatigue and the group attribution result is individual isolation, a yellow alert is generated, and mandatory leave and retesting on a later date are implemented; when the time attribution result is temporary fatigue and the group attribution result is training deficiency contagion, an orange alert is generated, and mandatory leave, online micro-course make-up learning, and delayed training for the work group are implemented; when the time attribution result is individual degeneration and the group attribution result is individual isolation, an orange alert is generated, and targeted retraining and a qualification observation period are implemented; when the time attribution result is individual degeneration and the group attribution result is negative demonstration contagion, a red alert is generated, and isolation retraining, interviews with work group leaders, and practical correction of work group errors are implemented; when the time attribution result is cumulative damage, a red alert is generated regardless of the group attribution result, and job reassignment assessment, special occupational health examination, and qualification freeze recommendations are implemented; when the time attribution result is hidden overwork and the group attribution result is external change contagion, an orange alert is generated, and reduced workload scheduling and joint training on new equipment are implemented. The attribution fusion decision-making unit also sets up an intervention conflict resolution mechanism. When mandatory leave conflicts with the training task time of the team, the time order is adjusted so that leave takes priority and training is postponed to the retest.

[0013] Preferably, the tiered intervention and closed-loop verification module includes a tiered intervention execution unit, used to push intervention instructions to employee mobile terminals, team management terminals, and the mine safety management system in a tiered manner through a distributed terminal access device, and to monitor the execution status: Yellow alert intervention: Push rest notification to employee mobile terminal, send a confirmation message for delayed assessment filing to team management terminal, employee scan code to confirm rest status on distributed terminal, system timer and automatically unlock retest entry after expiration; if the employee's well positioning card is detected to be activated during the rest period, send an intervention execution failure alarm to the safety supervision department. Orange alert intervention: Generate personalized learning paths, and employees complete micro-lesson learning and embedded micro-assessments on distributed terminals. When the learning completion rate is lower than the completion rate threshold, the qualification for formal retesting is locked. Red alert intervention: Send a qualification freeze recommendation to the mine safety management system and synchronize it to the location card permission system to restrict the employee from entering the critical work area; the team leader confirms the interview task on the management terminal, and the system records the interview duration and content summary. Team-level intervention: Team leaders confirm the team's health rectification list item by item on the management terminal, and the system randomly selects team members for surprise retesting to verify the rectification effect.

[0014] Preferably, the tiered intervention and closed-loop verification module further includes a differentiated closed-loop verification unit for performing differentiated verification and account cancellation management: Temporary fatigue type: The case is closed once the score of the first retest after the rest of the shift returns to the dynamic baseline. Individual deterioration type: After targeted retraining, two consecutive formal assessments are normal and there are no abnormalities in job behavior at the window for seven days, the account will be cancelled; Cumulative injury type: The new position will be removed from the list after three consecutive normal performance evaluations and passing the occupational health examination. Training defect contagion: After the team resets and retrains, the abnormal clustering index of the team returns to zero and the original abnormal employee passes the individual retest, and the team is removed from the list. Negative demonstration contagion: The team is removed from the list once the team passes the retest after corrective practice and the homogenization rate of underground collaborative behavior drops below the homogenization threshold. Blind spot risk: The account will be removed once the specific practical assessment is passed and the assessment for this knowledge point is normal within the following three months. The differentiated closed-loop verification unit feeds back the verification results to the individual abnormality screening module and the virtual-real coupled verification module according to the categories of true positive, false positive, false negative and true negative, so as to iteratively optimize their baseline parameters and verification thresholds.

[0015] Preferably, the model evolution iteration module includes an accident backtracking and association unit, which is used to automatically retrieve the early warning records of the accident area and related positions within the past 30 days within 24 hours after a safety accident or near-accident occurs in the mine. If the early warning system has marked the relevant employee, then trace the employee's complete intervention chain, check the accuracy of the attribution, the status of the intervention plan generation, the status of management confirmation, the status of the employee's execution, and the status of the location card permission freeze at each node, locate the intervention breakpoint, and generate a management accountability chain report. If the early warning system fails to flag relevant employees, the root causes of the missed reports will be analyzed, including excessively broad baseline settings, misjudgment of virtual-real coupling verification as state noise, or missing data sources. A model defect diagnosis report will be output and an emergency model iteration will be triggered.

[0016] The technical effects and advantages of this invention are as follows: (1) This invention aggregates assessment process data, job operation behavior data, production rhythm data and team group data through the data aggregation and alignment module, and performs time sequence alignment and feature extraction based on a unified time benchmark. Combined with the dynamic baseline construction and three-dimensional anomaly detection of the individual anomaly screening module, and the virtual-real coupling verification module to verify the coupling of assessment scenario data and job operation behavior data, it solves the problems of static assessment and virtual-real separation in the prior art, realizes the dynamic perception of employee quality risk and the effective identification of real risk, and improves the accuracy and reliability of assessment results. (2) This invention uses the time-rhythm coupling attribution unit and the work group attribution unit of the multidimensional coupling attribution module to conduct source analysis on employee abnormalities in terms of time dimension (temporary fatigue type, cumulative damage type, individual degeneration type, hidden overwork type) and group dimension (training defect contagion, negative demonstration contagion, external change contagion) respectively. It also uses the attribution fusion decision unit to perform cross mapping, which solves the problem of singular attribution in the prior art, realizes the accurate determination of the nature of individual abnormalities and the early identification of risk contagion in work groups, and provides accurate decision-making basis for differentiated intervention. (3) The present invention uses the hierarchical intervention execution unit of the hierarchical intervention and closed-loop verification module to push intervention instructions to the distributed terminal and the mine safety management system according to the comprehensive risk profile and monitor the execution status. At the same time, the differential closed-loop verification unit sets different verification cycles and cancellation conditions for different attribution types and feeds back the verification results to the individual anomaly screening module and the virtual-real coupling verification module to iteratively optimize its parameters. This solves the problems of formalized intervention and broken verification in the prior art, ensures the effective implementation of intervention instructions and full control of the execution status, and avoids the formalism of system suggestions but no execution on site. (4) The present invention uses the accident backtracking and association unit of the model evolution iteration module to automatically retrieve the warning records of related positions and trace the complete intervention link to locate the breakpoint after the accident occurs. At the same time, it diagnoses the root cause of the missed report and triggers the emergency iteration of the model, which solves the problem of evolutionary missingness in the prior art, realizes the data connection and linkage evolution of the quality management system and the mine safety management system, and forms a technical closed loop of warning-intervention-verification-evolution. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic block diagram of the system of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention provides, for example Figure 1 The diagram illustrates a distributed self-service management system for coal mine employee competence, comprising a server and distributed terminal access devices connected to the server via wired or wireless network communication. The distributed terminal access devices are configured to support coal mine employees' self-service access to the server. In this embodiment, the server is deployed at the mine's surface dispatch center, employing a dual-machine hot standby architecture. The distributed terminal access devices include an explosion-proof touchscreen all-in-one machine deployed in the mine entrance attendance room, desktop computers in the team meeting rooms, smartphones held by individual employees (accessing via underground Wi-Fi or surface 4G / 5G networks), and emergency tablets in refuge chambers. The server is configured to execute an intelligent early warning and tiered intervention method for competence risks, which is collaboratively implemented by the following functional modules: a data aggregation and alignment module, an individual anomaly initial screening module, a virtual-real coupling verification module, a multi-dimensional coupling attribution module, a tiered intervention and closed-loop verification module, and a model evolution iteration module.

[0020] The data aggregation and alignment module is configured to acquire multi-source heterogeneous data of employees from the distributed terminal access device and the existing information system of the mine. In this embodiment, the multi-source heterogeneous data includes: (1) assessment process data collected by the distributed terminal access device, including answer sequence, dwell time of each question, frequency of option switching, touch screen coordinate trajectory, correct answer rate of audio interactive questions, decision delay of VR emergency drill and correct operation steps; (2) job operation behavior data collected by the mine personnel positioning and equipment monitoring system (such as KJ251A type personnel positioning system), including the time of employees entering / leaving the mine, number of steps underground, dwell time in key areas, reporting records of hidden danger photos of smart mine lamps, and logs of equipment operation panels; (3) production rhythm data collected by the mine production scheduling system, including the type of production events such as relocation, crossing faults, gas extraction, and centralized maintenance, the code of the affected area, and the start and end timestamps of the plan; (4) team group data collected by the employee management database, including team code, team member list, recent training records, and underground collaborative positioning trajectory.

[0021] The data aggregation and alignment module performs time-series alignment and feature extraction on multi-source heterogeneous data based on a unified time benchmark. In this embodiment, the unified time benchmark adopts the server's NTP network clock (accuracy ±10ms), and the distributed terminal access device automatically performs clock calibration with the server at 03:00 every day. The time-series alignment uses the end time T0 of the assessment as the anchor point: Forward time-series features are extracted from the assessment process data, with a time-series granularity of seconds and a window length equal to the actual duration of the assessment (e.g., 25 minutes); For job operation behavior data, a first sliding window feature is extracted with T0 as the endpoint, with a time-series granularity of shifts (8 or 12 hours per shift) and a window length of 7 days (i.e., including the day of T0 and the preceding 6 shifts); For production rhythm data, a second sliding window feature is extracted centered on T0, with a time-series granularity of events and a window length of [T0-72 hours, T0+24 hours] to capture the production load impact of 3 shifts before and after the assessment and 1 shift after the assessment; For team group data, a team rolling monitoring window is established centered on T0, with a time-series granularity of days and a window length of 7 days.

[0022] For data missing situations, this embodiment adopts the following cleaning strategy: if the data missing rate in the assessment process is less than 5%, linear interpolation is used to complete it; if the missing rate is greater than or equal to 5% but less than 20%, it is marked as data quality degradation and the weight coefficient of this assessment is reduced to 60% in subsequent modules; if the missing rate is greater than or equal to 20%, the assessment is marked as invalid data and will not be included in the initial screening of individual anomalies. For job operation behavior data, if the location data for a certain shift is missing, it is filled with the average of the same shift in the employee's history; if three consecutive shifts are missing, a data source missing alarm is triggered, and the confidence weight of the employee's job operation behavior data is reduced in the virtual-real coupling verification module.

[0023] The individual anomaly screening module performs initial screening of individual quality risks based on a feature dataset. In this embodiment, the individual anomaly screening module includes a dynamic baseline construction unit, a three-dimensional anomaly detection unit, and a label generation unit.

[0024] The dynamic baseline construction unit establishes a dynamic baseline library stratified by job position, length of service, and skill level. Table 1 shows some examples of stratified baselines in this embodiment: Table 1. Examples of Dynamic Baseline Library Hierarchy

[0025] Newly hired employees will use a baseline replacement based on the same job position and length of service during their first month's performance evaluation period (defined as participating in evaluations for 30 consecutive days or accumulating 4 evaluations, whichever comes first). After the first month's evaluation period, a dynamic baseline based on the individual's data from the past 30 days will be used. The dynamic baseline uses a rolling update mechanism, incorporating the employee's latest valid evaluation scores from the past 30 days into the current baseline with the first weight, while the original dynamic baseline is retained with the remaining weight. The two are then weighted and merged to form the updated dynamic baseline. In this embodiment, the first weight is initially set at 30%, and is dynamically adjusted based on the false positive rate from subsequent closed-loop verification feedback. When the false positive rate increases, the first weight is reduced to mitigate baseline fluctuations; when the false positive rate decreases, the first weight is appropriately increased to enhance baseline sensitivity.

[0026] The 3D anomaly detection unit performs the following detections: (1) Vertical anomaly detection: Calculate the absolute difference between the current assessment score and the individual's dynamic baseline, and divide this difference by the fluctuation range of the individual's scores over the past 30 days to obtain the degree of vertical deviation. If the degree of vertical deviation is greater than twice the fluctuation range, it is judged as a vertical anomaly; (2) Horizontal anomaly detection: Calculate the absolute difference between the current assessment score and the average score of the same position and seniority group, and divide this difference by the fluctuation range of the group scores to obtain the degree of horizontal deviation. If the degree of horizontal deviation is greater than 1.5 times the fluctuation range of the group scores, it is judged as a horizontal anomaly; (3) Detection of abnormal behavior: Monitoring characteristic degradation patterns during the answering process. In this embodiment, if the skipping rate of safety procedure questions (such as the first step of gas over-limit handling) is greater than 30%, or the frequency of option switching decreases by more than 50% compared with the employee's historical average (determined as abandonment of answering questions), or the decision delay in the face of gas alarm scenarios in VR drills is extended by more than 50% compared with the individual baseline, then it is determined as abnormal behavior.

[0027] The label generation unit generates individual anomaly labels based on the 3D anomaly detection results. The data structure of the individual anomaly label is: Label={anomaly type, anomaly intensity, anomaly knowledge module}. The anomaly type is a combination of vertical, horizontal, and behavioral codes (e.g., vertical + behavioral code is LB). The anomaly intensity is determined according to the following rules: a single-dimensional anomaly with deviation between the threshold and the threshold plus 0.5 times the fluctuation range is considered mild; a single-dimensional anomaly with deviation exceeding the threshold plus 0.5 times the fluctuation range, or any two combined anomalies, is considered moderate; all three are anomalies, or any dimension with deviation exceeding the threshold plus 1.5 times the fluctuation range, is considered severe. The anomaly knowledge module automatically extracts information through clustering of incorrect answers (e.g., self-rescue device cylinder pressure verification, roof support procedures, etc.).

[0028] The virtual-real coupling verification module performs virtual-real coupling verification on individual anomaly labels. In this embodiment, the virtual-real coupling verification module includes a trigger determination unit, a coupling determination unit, a noise source tracing unit, and a blind zone handling unit.

[0029] The triggering judgment unit only activates the job operation behavior database for coupled verification when the anomaly intensity of an individual anomaly label reaches moderate or higher. If the anomaly intensity is mild, it is directly marked as an observation period, and subsequent coupled verification is not triggered to reduce the system's computational load.

[0030] The coupling determination unit performs the following judgment based on the coupling degree between the assessment scenario data and the job operation behavior data. In this embodiment, the coupling degree is determined through multi-dimensional weighted fusion, specifically: the performance trend matching degree is assigned a first fusion weight, the key behavior matching degree is assigned a second fusion weight, and the procedure execution matching degree is assigned a third fusion weight. The three are weighted and superimposed to obtain the final weighted matching degree. Among them, the performance trend matching degree is defined as the correlation between the slope of the moving average change of the performance of the last three assessments and the slope of the moving average change of the comprehensive score of job operation behavior in the last seven days; the key behavior matching degree is defined as the activity deviation of the job behavior corresponding to the abnormal knowledge module in the assessment; and the procedure execution matching degree is defined as the changing trend of the accuracy rate of the relevant parameter settings in the equipment operation panel.

[0031] The judgment criteria are as follows: If the weighted matching degree is greater than or equal to 0.6 and the performance trend and the job behavior trend are both declining in the same direction, it is judged as strong coupling, and a real risk label is generated. If the weighted matching degree is less than 0.3 and the job behavior trend is stable or rising, it is judged as weak coupling and a state noise label is generated. If the weighted matching degree is greater than or equal to 0.6, but the performance trend is stable or rising while the job behavior trend is declining, it is judged as reverse coupling, and a blind spot risk label is generated.

[0032] When generating status noise labels, the noise source tracing unit retrieves the terminal environment logs from this assessment and performs the following refined source tracing: Strong light accidental touch noise: When the light intensity is greater than 800 lux (read by the terminal's front-facing light sensor) and the operation is a touch screen, the straight-line distance between the actual touch point position and the center position of the target option is further calculated, and this distance is compared with the option radius to obtain the relative deviation ratio. If the relative deviation ratio is greater than 15%, it is judged as strong light accidental touch noise. A typical scenario is the midday sun in a wellhead attendance room during summer, when the explosion-proof touch screen is directly exposed to sunlight. Environmental interference noise: When the microphone noise floor is greater than 60dB (equivalent sound pressure level read from the terminal microphone) and the current question type is an audio interactive question (such as identifying the beeping frequency of a gas alarm by sound), it is judged as environmental interference noise. A typical scenario is multiple people talking during a pre-shift meeting in the district meeting room; Physiological fatigue noise: When the employee's heart rate variability coefficient (uploaded by the heart rate module built into the smart miner's lamp or the employee's wearable device) is lower than the third threshold (set to 20ms in this embodiment) and the duration of the mine descent is greater than the fourth threshold (set to 8 hours in this embodiment), it is determined to be physiological fatigue noise; Emotional stress noise: When none of the above three conditions are met, it is judged as emotional stress noise. At this time, the server pushes the Psychological Resilience Self-Assessment Scale to the employee's mobile terminal (example question: In the past week, have you had difficulty concentrating due to family matters?, using a Likert 5-point rating). If the total score is greater than 12 points, it is marked as emotional stress and it is recommended to retake the test on another day.

[0033] When generating blind spot risk labels, the blind spot handling unit semantically matches the abnormal features of job operations with the assessment question bank. In this embodiment, a knowledge graph mapping job operations to assessment knowledge points is constructed. Nodes include types of operational anomalies such as equipment parameter setting deviations, missed inspection routes, and non-standard descriptions of hazards, as well as assessment knowledge points such as self-rescue device cylinder pressure verification steps and gas power-off device power restoration procedures. Semantic matching employs a dual judgment of keyword co-occurrence and semantic similarity: First, verb-noun phrases (such as settings, parameters, deviations) are extracted from the job operation anomaly log and their text similarity is calculated with the knowledge point labels of each question in the assessment question bank. If the maximum similarity is less than 0.4, it is determined that the operational anomaly has no corresponding assessment point in the assessment question bank, is marked as an assessment blind spot, and triggers targeted question bank supplementation (added to the database after administrator review) and a special practical assessment task (pushed to the VR simulation module on the employee's terminal).

[0034] For example, employee Zhang (a gas inspector with 2 years of service and an intermediate-level worker) completed a VR assessment of wearing a self-rescue device in the mine entrance attendance room after coming up from the mine on May 10, 2026. His score dropped to 62 points from his personal dynamic baseline (85 points). The vertical deviation exceeded twice the standard deviation threshold, and the horizontal deviation also exceeded 1.5 times the group standard deviation threshold. Furthermore, his behavior was abnormal (35% skipping rate of safety procedure questions), reaching a moderate level of abnormality, triggering coupling verification. His work operation behavior data for the past 7 days showed that the frequency of hazard reporting decreased from the historical average of 5.2 times / shift to 1 time / shift, there were 0 photos of hazard-related issues related to the self-rescue device in the smart mine lamp, and there were no abnormalities on the equipment operation panel. After multi-dimensional weighted fusion calculation, the weighted matching degree was far below the strong coupling judgment standard, falling within the weak coupling range, and was judged as weakly coupled, generating a state noise label. Further review of the terminal environment logs revealed that the assessment occurred at 12:30 PM, with a light intensity of 0 lux, a relative deviation rate of 22%, and Zhang's shift duration in the well was 9.5 hours, with a heart rate coefficient of variation of 18 ms. Based on this, the system determined that there was a combined effect of strong light-induced accidental exposure noise and physiological fatigue noise. The system flagged this assessment as state noise – composite noise, and sent Zhang a notification suggesting a 4-hour rest period followed by a retest after 4:00 PM. The system also sent a record to the team management terminal stating that Zhang's assessment was invalid and requesting a make-up test.

[0035] The multidimensional coupled attribution module performs multidimensional coupled attribution analysis based on risk authenticity labels. In this embodiment, the multidimensional coupled attribution module includes a time-rhythm coupled attribution unit, a work group attribution unit, and an attribution fusion decision unit.

[0036] The time-rhythm coupling attribution unit is integrated with the mine production scheduling system to establish a production event calendar. The data structure of the production event calendar is: Event={Event Type, Affected Area Code, Planned Start and End Timestamps, Actual Start and End Timestamps, List of Associated Positions}. For example, the production event from May 5th to May 15th, 2026 is the relocation and overhaul of a fully mechanized mining face, with the affected area code ZC-001, and associated positions including fully mechanized mining face support workers, gas inspectors, and conveyor belt drivers.

[0037] The time-rhythm coupled attribution unit constructs an individual production load index by combining the individual employee's shift scheduling rhythm. In this embodiment, the individual production load index is determined by a weighted fusion of three factors: downhole duration, downhole activity intensity, and night shift frequency. Downhole duration has the dominant weight, while downhole activity intensity and night shift frequency each have auxiliary weights. The weighted sum of these three factors yields the individual production load index. The load threshold is set to 0.75 in this embodiment.

[0038] When the risk authenticity label is identified as genuine risk, the time-rhythm coupled attribution unit performs the following judgment: If an individual's production load index exceeds the load threshold and the current time point is within the production event window, and the employee has experienced anomalies more than or equal to 3 times under similar load conditions in the past 6 months (similar load is defined as a load index deviation within ±10%), then it is determined to be a cumulative damage type. If an individual's productivity load index exceeds the load threshold and the current time point is within the production event window, but the number of similar load anomalies is less than 3 in the past 6 months, it is determined to be a temporary fatigue type. If an individual's productivity load index does not exceed the load threshold and the current time is far from the production event window (more than 72 hours from the end of the most recent event), then the individual is classified as degenerate. If an individual's productivity load index exceeds the load threshold and the current time point is far from the production event window, it is determined to be a case of hidden overwork.

[0039] The work group attribution unit establishes a seven-day rolling monitoring window centered on the end time T0 of the assessment, and calculates the anomaly clustering index. In this embodiment, the similarity of common blind spots is determined by set comparison, that is, by counting the number of knowledge points that appear together in the abnormal knowledge modules of multiple abnormal employees, and dividing it by the total number of knowledge points covered by the abnormal knowledge modules of all abnormal employees, the overlap ratio of common blind spots is obtained. The anomaly clustering index is the product of the number of abnormal employees and the overlap ratio of common blind spots.

[0040] For example, if 3 out of 5 members in a certain work group exhibit abnormalities within a seven-day window, the abnormality knowledge modules are as follows: Employee A: {Self-rescue device cylinder pressure calibration, gas disconnect device power restoration operation, roof support procedures} Employee B: {Self-rescue device cylinder pressure calibration, gas shut-off device power restoration operation, disaster evacuation route identification} Employee C: {Self-rescue device cylinder pressure calibration, gas disconnect device power restoration operation, CPR procedures} The common knowledge points are {self-rescue device cylinder pressure verification, gas power cut-off device power restoration operation}, and the total number of knowledge points is {self-rescue device cylinder pressure verification, gas power cut-off device power restoration operation, roof support procedures, disaster evacuation route identification, cardiopulmonary resuscitation steps}. The overlap ratio of common blind spots is 2 divided by 5, which is 40%. The number of abnormal employees is 3, and the abnormal clustering index is 3 multiplied by 40%, which is 1.2. The clustering threshold is set to 1.0 in this embodiment, therefore, this team triggers group attribution.

[0041] When the abnormal clustering index is greater than or equal to the clustering threshold, the work group attribution unit initiates a three-level tracing process: Training Input Tracing: Retrieve the training records of the work group for the past 30 days and extract the knowledge point tag set of the training courses. Count the number of knowledge points coexisting in the common blind spot set and the training course knowledge point tag set, divide this number by the total number of knowledge points covered by both sets, and obtain the training content matching ratio. If the training content matching ratio is less than 60%, it is considered a training defect contagion. In the above example, if the training courses in the past 30 days did not include self-rescue device cylinder pressure calibration and gas disconnector power restoration operation, the matching ratio is 0, and it is considered a training defect contagion. Team behavior tracing: Analyze underground collaboration data. Count the number of hazard photos with highly similar equipment scenes and descriptions within the team, divide by the total number of hazards reported by the team, and obtain the hazard photo homogeneity ratio. If the homogeneity ratio is higher than 50%, and the location trajectory shows a pairing pattern between experienced and new employees (defined as: location distance less than 5 meters within 2 hours and occurring 3 or more times consecutively), it is judged as a negative demonstration of contagion; External environment tracing: Search for records of mine equipment changes and procedure adjustments. If there are equipment changes or procedure adjustments within the past 30 days, and the content of the changes is strongly correlated with common blind spots (keyword matching degree greater than 0.7), it is determined to be an external change contagion.

[0042] The attribution fusion decision-making unit constructs an attribution fusion decision matrix, cross-mapping the attribution results of the time-rhythm coupled attribution unit with the attribution results of the work group attribution unit to generate a comprehensive risk level and precise intervention plan. Table 2 shows the attribution fusion decision matrix in this embodiment: Table 2 Attribution Fusion Decision Matrix

[0043] The attribution fusion decision-making unit also includes an intervention conflict resolution mechanism. When an employee simultaneously triggers mandatory leave (due to temporary fatigue or cumulative damage time attribution) and a team training task (due to training deficiency contagion group attribution), the system performs conflict detection: calculating the proportion of overlap between the leave period and the training period to the total training time. If the overlap proportion is greater than 50%, leave is prioritized, and the training is postponed to the next test date, and the employee's training appointment time is automatically updated in the system.

[0044] The tiered intervention and closed-loop verification module executes tiered interventions based on a comprehensive risk profile and performs differentiated closed-loop verification of the intervention effects. In this embodiment, the tiered intervention and closed-loop verification module includes a tiered intervention execution unit and a differentiated closed-loop verification unit.

[0045] The tiered intervention execution unit pushes intervention instructions to employee mobile terminals, team management terminals, and the mine safety management system through distributed terminal access devices, and monitors the execution status.

[0046] The specific implementation methods for interventions at all levels are as follows: Yellow alert intervention: A rest notification (including suggested rest duration, automatically calculated by the system based on the individual's production load index; a 4-hour rest is recommended when the load index is between 0.75 and 1.0, and an 8-hour rest is recommended when the load index is greater than 1.0) is sent to the team management terminal, indicating that the postponement assessment registration is pending confirmation. Employees scan a code on the distributed terminal to confirm their rest status; the system keeps track of the time and automatically unlocks the retest entry upon expiration. If the employee's mine entry positioning card is detected to be activated during the rest period (achieved by polling the mine personnel positioning system's API interface every 5 minutes), an intervention execution failure alarm is sent to the safety supervision department. The alarm information includes the employee's name, card number, activation time, and activation area code. Orange alert intervention: Generate a personalized learning path. The learning path is automatically arranged by the system based on the abnormal knowledge module, including 3-5 micro-lesson videos (5-8 minutes each) and embedded micro-assessments (2-3 key step confirmation questions pop up after each video). Employees complete the learning on distributed terminals, and the system records the video viewing completion rate (calculated by dividing the actual playback time by the total video length, excluding fast-forwarding). When the completion rate is below 80%, the employee's eligibility for a formal retest is locked, and a prompt is sent to the employee requesting them to complete the learning before applying for a retest. Red Alert Intervention: A qualification freeze recommendation is sent to the mine safety management system and synchronized to the location card access control system via a standard interface, restricting the employee's access to critical work areas (such as fully mechanized mining faces, tunneling heads, and gas extraction roadways). Upon receiving the freeze order, the location card access control system will trigger an audible and visual alarm and refuse passage the next time the employee attempts to pass through the critical area access control. The team leader confirms the interview task on the management terminal, and the system records the interview duration (calculated using the management terminal's sign-in / sign-out timestamps) and a summary of the interview content (stored after the team leader's voice input is converted to text). Team-level intervention: The team leader confirms the team's health rectification checklist item by item on the management terminal (example checklist items: completed reiteration of the mentoring operation procedure, completed special training on common blind spots, completed quality review of hazard reporting). The system randomly selects 2 team members for a surprise retest (without prior notice, the retest questions are randomly selected from the common blind spots). If both members pass, the team's rectification is deemed effective.

[0047] The differentiated closed-loop verification unit performs differentiated verification and account cancellation management for different attribution types. Table 3 shows the differentiated account cancellation conditions in this embodiment: Table 3 Differentiated Closed-Loop Validation and Account Closure Conditions

[0048] The differentiated closed-loop verification unit classifies and feeds back the verification results as true positive, false positive, false negative, and true negative. In this embodiment, the four-category judgment logic is as follows: True positive: The system alert indicates a real risk, and subsequent verification (retesting, job behavior tracking, or incident correlation) confirms that the employee does indeed have a decline in ability or risky behavior; False positive: The system alerted that there was a real risk, but subsequent verification showed that the employee passed the retest and his / her job behavior was normal. The reason was found to be that the baseline setting was too wide or the terminal environmental noise was not completely eliminated. False negative: The system did not issue a warning or the warning was status noise, but the employee's subsequent job behavior became abnormal or he was involved in a safety accident. Retrospective investigation revealed that the cause was that the virtual-real coupling verification threshold was too high or the data source was missing. True negative: The system did not issue a warning or the warning was status noise, and subsequent verification confirmed that the employee's status was normal.

[0049] The differentiated closed-loop verification unit feeds back the above four-category classification results to the individual anomaly screening module and the virtual-real coupling verification module to iteratively optimize their parameters. In this embodiment, the parameter iteration rules are as follows: If the false positive rate exceeds 15%, the baseline update rate of the individual abnormality screening module will be attenuated and adjusted. Specifically, the current update rate will be multiplied by the false positive correction coefficient, which is determined based on the false positive rate. The higher the false positive rate, the lower the correction coefficient, thereby reducing the baseline update rate and relaxing the deviation judgment threshold of the dynamic baseline (from 2.0 times the fluctuation range to 1.8 times the fluctuation range) to reduce oversensitivity. If the false negative rate exceeds 10%, the coupling degree judgment criteria of the virtual-real coupling verification module will be tightened (the strong coupling judgment threshold will be reduced from 0.6 to 0.5, and the weak coupling judgment threshold will be reduced from 0.3 to 0.2), and the judgment weight of state noise will be reduced. If the blind spot detection rate (the proportion of false negatives caused by blind spots in the assessment) exceeds 20%, the question bank supplementation process will be triggered, and the assessment weight of that knowledge point will be increased by 30%.

[0050] The model evolution iteration module performs model evolution iteration based on the results of closed-loop verification and the correlation of incident backtracking. In this embodiment, the model evolution iteration module includes an incident backtracking correlation unit.

[0051] When a safety accident or near miss occurs in a mine, the accident retrospective correlation unit automatically retrieves warning records for the accident area and related positions within the past 30 days within 24 hours of the accident. Automatic retrieval is achieved by subscribing to the accident reporting message queue of the mine safety management system. Once an accident reporting message (including the accident area code, a list of related positions, and the accident level) is received, the retrieval task is immediately triggered.

[0052] If the early warning system has already flagged a relevant employee, the entire intervention chain for that employee is traced. The status nodes in the intervention chain include: attribution accuracy (whether time-based and group-based attributions are correct), intervention plan generation status (whether a JSON instruction has been generated), management confirmation status (whether the team leader clicked confirmation on the management terminal), employee execution status (whether leave of absence, retraining, or interviews have been completed), and location card permission freeze status (whether the location card permission system returned a successful freeze receipt). The system checks the status logs node by node, identifies the first node with a failure or timeout status as the intervention breakpoint, and generates a management accountability chain report. The report includes: breakpoint type, responsible management role, suggested corrective measures, and related system interface log fragments.

[0053] If the early warning system does not flag the relevant employees, then the root cause diagnosis of missed reports will be initiated. The diagnostic logic is as follows: If the deviation of an employee's historical performance evaluation from their individual dynamic baseline is far below the normal fluctuation range and does not reach the minimum deviation threshold for triggering a longitudinal anomaly, it is determined that the baseline setting is too wide. A model defect diagnosis report is output, and it is recommended that the baseline construction for this job group be changed from the median to the lower quartile to improve sensitivity. If the employee's most recent anomaly is marked as state noise by the virtual-real coupling verification module, but the employee's behavior at the same time has shown signs of degradation, it is determined to be a misjudgment of the virtual-real coupling verification. It is recommended to lower the weak coupling judgment threshold from 0.3 to 0.2 and increase the weight of physiological fatigue in noise source tracing. If the employee is a member of an outsourced team and there is no record of their location data in the mine personnel positioning system, it is determined that the data source is missing. It is recommended to add an outsourced personnel data access interface in the data aggregation and alignment module and set a missing data marking strategy.

[0054] The model evolution iteration module supports two iteration modes: Online hot update: Applicable to threshold parameter adjustment (such as baseline update rate, coupling degree judgment threshold, aggregation threshold), the update process does not restart the server, and is distributed to each functional module in real time through the configuration center; Offline retraining: Suitable for model structure optimization (such as weight adjustment for 3D anomaly detection and reconstruction of the mapping relationship of attribution fusion decision matrix), executed from 02:00 to 04:00 on the first Sunday of each month, with training data being the full validation results of the previous month.

[0055] Take employee Li (a support worker at a fully mechanized mining face, with 4 years of service and a senior technician qualification) as an example: Step S1: On May 12, 2026, after coming up from the well on the morning shift, Li completed the standardized operation assessment for his position on the computer terminal in the team's meeting room. His score was 78 points, which was lower than his personal dynamic baseline (90 points). The vertical deviation exceeded twice the fluctuation amplitude threshold, and the horizontal deviation also exceeded 1.5 times the group fluctuation amplitude threshold. In addition, his behavior was abnormal (the skipping rate of the safety procedure questions was 35%), and the abnormality intensity was moderate. Step S2: Trigger virtual-real coupling verification. Data on the employee's operational behavior over the past 7 days shows that the frequency of hazard reporting has decreased from a historical average of 6 times / shift to 2 times / shift, and there was one instance of incorrect setting of fully mechanized mining face support parameters. Through multi-dimensional weighted fusion calculation, the weighted matching degree is within the strong coupling range, thus it is determined to be strongly coupled, and a real risk label is generated. Step S3: Time-Rhythm Coupled Attribution. The individual's production load index exceeded the load threshold during the fault-crossing production event window of the fully mechanized mining face (May 10th to May 20th). The number of abnormal occurrences under similar load conditions within the past 6 months was 4 (1 each in February, March, April, and May), classifying it as a cumulative damage type. Step S4: Group Attribution. Two individuals within the work group exhibited abnormalities within the past seven days. The common blind spots were {initial hydraulic support force setting, roof separation meter reading identification}, with a 50% overlap rate. The number of abnormal employees was 2, and the abnormal clustering index exceeded the clustering threshold. Training input tracing showed that training in the past 30 days did not cover the above knowledge points (training content matching rate was 0), indicating a training deficiency contagion. Step S5: Attribution Integration Decision. A combination of cumulative impairment and training defect contagion is used to generate a red alert based on Table 2, prompting recommendations for job reassignment assessment, specialized occupational health examination, and qualification freeze. Step S6: Implementation of tiered intervention. The system pushes a red alert notification to Li's mobile terminal, sends a qualification freeze recommendation to the mine safety management system, and synchronizes with the location card access control system to restrict his access to the fully mechanized mining area. Team leader Wang confirms receipt of the interview task on the management terminal; Step S7: Differentiated Closed-Loop Verification. Li underwent an occupational health examination (lumbar strain, unsuitable for high-intensity support work) and was transferred to the ground equipment maintenance position. His new position resulted in three consecutive performance evaluations of 88, 91, and 89 points, all within the normal range. His occupational health examination was re-examined, and the system closed the registration. Step S8: Model Evolution. This case was verified as a true positive, the system recorded the attribution accuracy (correct time attribution, correct population attribution), the intervention link remained unbroken, and the model parameters remained unchanged.

[0056] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A self-service management system for the professional development of coal mine workers that can be distributed, characterized in that: include: A server, and a distributed terminal access device connected to the server via a wired or wireless network, the distributed terminal access device being configured to support coal mine employees' self-service access to the server. The data aggregation and alignment module is used to obtain multi-source heterogeneous data of employees from distributed terminal access devices and existing information systems in the mine, and to perform time-series alignment and feature extraction on the multi-source heterogeneous data based on a unified time benchmark to generate a feature dataset. The individual anomaly screening module is used to perform initial screening of individual quality risks based on feature datasets and generate individual anomaly labels. The virtual-real coupling verification module is used to perform virtual-real coupling verification on individual abnormal labels, and generates risk authenticity labels based on the coupling degree between assessment scenario data and job operation behavior data; The multidimensional coupled attribution module is used to perform multidimensional coupled attribution analysis based on risk authenticity labels. The multidimensional coupled attribution analysis includes time-rhythm coupled attribution, work group clustering scanning and organizational attribution, generating a comprehensive risk profile. The graded intervention and closed-loop verification module is used to implement graded interventions based on the comprehensive risk profile and to perform differentiated closed-loop verification of the intervention effect. The verification results are fed back to the individual anomaly screening module and the virtual-real coupling verification module to iteratively optimize their parameters. The model evolution iteration module is used to perform model evolution iteration based on the results of closed-loop verification and the correlation of incident backtracking.

2. The self-service management system for the quality improvement of coal mine workers, which can be distributed according to claim 1, is characterized in that, Multi-source heterogeneous data includes assessment process data collected by distributed terminal access devices, job operation behavior data collected by mine personnel positioning and equipment monitoring systems, production rhythm data collected by mine production scheduling systems, and team group data collected by employee management databases. The data aggregation and alignment module uses the end time of the assessment as the anchor point to extract forward time-series features from the assessment process data, extracts first sliding window features with the end time of the assessment as the endpoint from the job operation behavior data, extracts second sliding window features with the end time of the assessment as the center from the production rhythm data, and establishes a rolling monitoring window for the work group data with the end time of the assessment as the center.

3. The self-service management system for the quality improvement of coal mine workers, which can be distributed according to claim 1, is characterized in that, The individual anomaly screening module includes: The dynamic baseline construction unit is used to establish a dynamic baseline library layered by job position, length of service, and skill level. New employees use the same job position and length of service group as a substitute baseline during the first month's assessment cycle. After the first month's assessment cycle ends, it switches to an individual dynamic baseline that is updated on a rolling basis based on the individual's data over the past thirty days. The three-dimensional anomaly detection unit is used to perform vertical anomaly detection, horizontal anomaly detection, and behavioral anomaly detection. Vertical anomaly detection is used to determine the degree of deviation of the current assessment score from the individual's dynamic baseline. Horizontal anomaly detection is used to determine the degree of deviation of the current assessment score from the average of the same position and the same length of service group. Behavioral anomaly detection is used to determine whether there are characteristic degradation patterns in the answering process. The tag generation unit is used to generate individual anomaly tags containing anomaly type, anomaly intensity, and anomaly knowledge modules based on the 3D anomaly detection results.

4. The self-service management system for the quality improvement of coal mine workers, which can be distributed according to claim 1, is characterized in that, The virtual-real coupling verification module includes: The trigger judgment unit is used to activate the job operation behavior database for coupled verification only when the abnormality intensity of an individual abnormal label reaches moderate or above. The coupling judgment unit is used to perform the following judgments based on the coupling degree between the assessment scenario data and the job operation behavior data: if the assessment degradation signal is strongly coupled with the job operation behavior degradation signal, a real risk label is generated; if the assessment degradation signal is weakly coupled with the job operation behavior normal signal, a state noise label is generated; if the assessment normal signal is reversely coupled with the job operation behavior abnormal signal, a blind zone risk label is generated. The noise source tracing unit is used to retrieve the terminal environment log of this assessment when a status noise label is generated. When the light intensity is greater than the first threshold and it is a touch screen operation, it is determined to be strong light accidental touch noise. When the ambient noise is greater than the second threshold and it is an audio interactive question, it is determined to be environmental interference noise. When the employee's heart rate variation coefficient is lower than the third threshold and the time spent in the well is greater than the fourth threshold, it is determined to be physiological fatigue noise. When none of the above conditions are met, it is determined to be emotional stress noise. The blind spot handling unit is used to semantically match the abnormal characteristics of job operations with the assessment question bank when a blind spot risk label is generated. If there is no corresponding assessment point, it is marked as an assessment blind spot and triggers targeted supplementation of the question bank and special practical assessment.

5. A self-service management system for the quality improvement of coal mine workers that can be distributed according to claim 4, characterized in that, The multidimensional coupled attribution module includes a time-rhythm coupled attribution unit, which is used to connect with the mine production scheduling system to establish a production event calendar and to construct an individual production load index by combining the individual shift schedule rhythm of employees. When the risk authenticity label is real risk, the time-rhythm coupled attribution unit performs the following judgment: if the individual production load index is greater than the load threshold and the current time point is within the production event window, and the number of times the employee has experienced abnormalities under similar load conditions in the past six months is greater than or equal to three, then it is judged as cumulative damage type; if it is less than three times, then it is judged as temporary fatigue type. If an individual's productivity load index is normal and the current time point is far from the production event window, it is determined to be an individual degenerative type; if an individual's productivity load index is greater than the load threshold and the current time point is far from the production event window, it is determined to be a latent overwork type.

6. A self-service management system for the quality improvement of coal mine workers, as described in claim 5, is characterized in that... The multidimensional coupling attribution module also includes a team group attribution unit, which is used to establish a seven-day rolling monitoring window for the team with the end of the assessment as the center, and calculate the abnormal clustering index. The abnormal clustering index is equal to the product of the number of abnormal people and the similarity of the common blind spot. When the abnormal clustering index is less than the clustering threshold, it is marked as an isolated event. When the abnormal clustering index is greater than or equal to the clustering threshold, a three-layer tracing mechanism is activated: Training input tracing, which retrieves recent training records for the work group; when the matching degree between the training content and the abnormal knowledge module is lower than the matching threshold, it is determined to be a training defect infection; Work group behavior tracing, which analyzes underground collaboration data; when the homogenization rate of hazard photos is higher than the homogenization threshold and there is a pairing trajectory between old and new employees, it is determined to be a negative demonstration infection; External environment tracing, which retrieves records of mine equipment changes and procedure adjustments; when the abnormal module is strongly correlated with recent changes, it is determined to be an external change infection.

7. A self-service management system for the quality improvement of coal mine workers that can be distributed according to claim 6, characterized in that, The multidimensional coupled attribution module also includes an attribution fusion decision unit, used to construct an attribution fusion decision matrix, cross-mapping the attribution results of the time-rhythm coupled attribution unit with the attribution results of the work group attribution unit to generate a comprehensive risk level and precise intervention plan: When the time attribution result is temporary fatigue and the group attribution result is individual isolation, a yellow alert is generated, and mandatory leave and retesting on a later date are implemented; when the time attribution result is temporary fatigue and the group attribution result is training deficiency contagion, an orange alert is generated, and mandatory leave, online micro-course make-up learning, and delayed training for the work group are implemented; when the time attribution result is individual degeneration and the group attribution result is individual isolation, an orange alert is generated, and targeted retraining and a qualification observation period are implemented; when the time attribution result is individual degeneration and the group attribution result is negative demonstration contagion, a red alert is generated, and isolation retraining, interviews with work group leaders, and practical correction of work group errors are implemented; when the time attribution result is cumulative damage, a red alert is generated regardless of the group attribution result, and job reassignment assessment, special occupational health examination, and qualification freeze recommendations are implemented; when the time attribution result is hidden overwork and the group attribution result is external change contagion, an orange alert is generated, and reduced workload scheduling and joint training on new equipment are implemented. The attribution fusion decision-making unit also sets up an intervention conflict resolution mechanism. When mandatory leave conflicts with the training task time of the team, the time order is adjusted so that leave takes priority and training is postponed to the retest.

8. A self-service management system for the quality of coal mine workers that can be distributed according to claim 7, characterized in that, The tiered intervention and closed-loop verification module includes a tiered intervention execution unit, used to push intervention commands tiered to employee mobile terminals, team management terminals, and the mine safety management system via a distributed terminal access device, and to monitor the execution status. Yellow alert intervention: Push rest notification to employee mobile terminal, send a confirmation message for delayed assessment filing to team management terminal, employee scan code to confirm rest status on distributed terminal, system timer and automatically unlock retest entry after expiration; if the employee's well positioning card is detected to be activated during the rest period, send an intervention execution failure alarm to the safety supervision department. Orange alert intervention: Generate personalized learning paths, and employees complete micro-lesson learning and embedded micro-assessments on distributed terminals. When the learning completion rate is lower than the completion rate threshold, the qualification for formal retesting is locked. Red alert intervention: Send a qualification freeze recommendation to the mine safety management system and synchronize it to the location card permission system to restrict the employee from entering the critical work area; the team leader confirms the interview task on the management terminal, and the system records the interview duration and content summary. Team-level intervention: Team leaders confirm the team's health rectification list item by item on the management terminal, and the system randomly selects team members for surprise retesting to verify the rectification effect.

9. A self-service management system for the quality improvement of coal mine workers that can be distributed according to claim 1, characterized in that, The tiered intervention and closed-loop verification module also includes a differentiated closed-loop verification unit, used to perform differentiated verification and account cancellation management: Temporary fatigue type: The case is closed once the score of the first retest after the rest of the shift returns to the dynamic baseline. Individual deterioration type: After targeted retraining, two consecutive formal assessments are normal and there are no abnormalities in job behavior at the window for seven days, the account will be cancelled; Cumulative injury type: The new position will be removed from the list after three consecutive normal performance evaluations and passing the occupational health examination. Training defect contagion: After the team resets and retrains, the abnormal clustering index of the team returns to zero and the original abnormal employee passes the individual retest, and the team is removed from the list. Negative demonstration contagion: The team is removed from the list once the team passes the retest after corrective practice and the homogenization rate of underground collaborative behavior drops below the homogenization threshold. Blind spot risk: The account will be removed once the specific practical assessment is passed and the assessment for this knowledge point is normal within the following three months. The differentiated closed-loop verification unit feeds back the verification results to the individual abnormality screening module and the virtual-real coupled verification module according to the categories of true positive, false positive, false negative and true negative, so as to iteratively optimize their baseline parameters and verification thresholds.

10. A self-service management system for the quality improvement of coal mine workers, which can be distributed according to claim 1, is characterized in that, The model evolution iteration module includes an accident backtracking and association unit, which is used to automatically retrieve the early warning records of the accident area and related positions within the past 30 days within 24 hours after a safety accident or near-accident occurs in the mine. If the early warning system has marked the relevant employee, then trace the employee's complete intervention chain, check the accuracy of the attribution, the status of the intervention plan generation, the status of management confirmation, the status of the employee's execution, and the status of the location card permission freeze at each node, locate the intervention breakpoint, and generate a management accountability chain report. If the early warning system fails to flag relevant employees, the root causes of the missed reports will be analyzed, including excessively broad baseline settings, misjudgment of virtual-real coupling verification as state noise, or missing data sources. A model defect diagnosis report will be output and an emergency model iteration will be triggered.

Images