A digital-based coal mine post training method and system

Abstract

The present application relates to the field of coal mine safety production and digital training, and particularly relates to a kind of coal mine post training method and system based on digitalization, the method includes the steps of collecting basic multi-source data, establishing data warehouse;Causal inference method is used to build causal association model, the causal effect of different training variables on safety performance is quantified;Extract the multi-dimensional safety risk characteristics of employees, after quantization, normalization, screening and dimension reduction, establish the employee safety risk characteristic library;Based on the characteristic library and historical accident data, a safety risk prediction model is built, the safety risk index of employees is calculated and the risk level is divided;According to the risk level, generate graded early warning information, and develop individualized intervention measures combined with employee risk characteristics.The present application establishes the causal association model of training data and safety performance data, realizes the quantitative evaluation of training effect, and builds a safety risk prediction model based on multi-dimensional data, identifies high-risk employees in advance and carries out precise intervention.

Description

Technical Field

[0001] This invention belongs to the field of coal mine safety production and digital training, specifically involving a digital coal mine job training method and system. Background Technology

[0002] As a high-risk industry, the coal mining sector directly impacts the safety of its employees and the sustainable development of enterprises. Safety training is a core component in improving employee safety awareness and reducing accident rates. However, current coal mining enterprises face numerous technical challenges in their safety training and management systems, severely hindering the improvement of safety management levels.

[0003] Current training effectiveness evaluations are mostly based on simple correlation analyses, which can only prove a link between training and safety performance, but cannot identify the causal relationship between the two. Furthermore, they cannot distinguish the impact of training factors on safety performance from non-training factors such as equipment, management, environment, and individuals, leading to distorted evaluation results and failing to provide a scientific basis for the rational allocation of training resources. At the same time, traditional training uses uniform content and formats, ignoring the heterogeneity of training effectiveness among employees of different ages, lengths of service, and skill levels, resulting in serious waste of training resources and insufficient targeting.

[0004] In terms of safety risk management, existing methods mainly focus on handling violations and accidents after the fact, lacking the ability to predict risks in advance and failing to identify high-risk employees in advance. Even if some companies have established risk assessment systems, they often rely on single indicators such as test scores and historical violation records, failing to comprehensively extract multi-dimensional safety risk characteristics such as employee learning behavior and job operations, resulting in low accuracy in risk identification and serious delays in early warning.

[0005] Furthermore, coal mining enterprises suffer from significant data silos across their various business systems, hindering cross-dimensional data integration and analysis of training, safety, equipment, and human resources data. Intervention measures are often implemented sequentially, resulting in low efficiency and a lack of real-time monitoring and quantitative evaluation of their effectiveness, making closed-loop management difficult. Model iteration and updates rely on manual operation, leading to high costs and poor timeliness. This makes it impossible to adapt to dynamic changes in the coal mine production environment, management systems, and employee structure, causing model performance to degrade rapidly over time and hindering long-term effective support for safety management. Summary of the Invention

[0006] The purpose of this invention is to provide a digital-based coal mine job training method and a digital-based coal mine job training system to solve the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A digital-based coal mine job training method includes the following steps: S1. Collect multi-source data on coal mine enterprise training, safety performance, equipment operation, management, and employee basics. After cleaning, completion, and standardization, establish a unified coal mine safety data warehouse with a star-shaped model architecture. S2. Classify and quantify four categories of non-training interference factors: equipment, management, environment, and individuals. Construct a control matrix for interference factors and eliminate multicollinearity. Calculate the net training effect for each employee through regression analysis. S3. Using training variables as treatment variables and net training effect as outcome variables, a causal association model is constructed using causal inference methods to quantify the causal effects of different training variables on safety performance; S4. Extract multi-dimensional safety risk characteristics of employees, and after quantification, normalization, screening and dimensionality reduction, establish an employee safety risk characteristic database; S5. Construct a safety risk prediction model based on the feature library and historical accident data, calculate the employee safety risk index and classify the risk level; S6. Generate graded early warning information based on risk levels, and formulate personalized intervention measures in combination with employee risk characteristics; S7. Regularly collect new data to update the data warehouse and feature library, and verify and iteratively optimize the causal relationship model and the safety risk prediction model.

[0008] Furthermore, in step S1, Z-score standardization is used for numerical data and one-hot encoding is used for categorical data; the data collection frequency is divided into real-time, daily, weekly and monthly according to type, and the data integrity is not less than 98%.

[0009] Furthermore, in step S2, the weights of each interfering factor are determined using the analytic hierarchy process (AHP), and a regression model is established with the comprehensive safety performance index as the dependent variable. The net training effect is obtained by subtracting the expected safety performance value from the actual safety performance value.

[0010] Furthermore, in step S3, the training variables include training content, training method, training duration, and training frequency; logistic regression is used to estimate propensity score, and nearest neighbor matching is used to match the treatment group with the control group to calculate the average treatment effect and heterogeneous treatment effect.

[0011] Furthermore, in step S4, the extracted safety risk features include four categories: learning behavior, exam results, historical violations, and job operations; the minimum-maximum normalization is used to unify the feature value range, and dimensionality reduction is performed through principal component analysis.

[0012] Furthermore, in step S5, the historical samples are divided into training and testing sets in a 7:3 ratio, and a prediction model is constructed using logistic regression with L1 regularization; the safety risk level is divided into four levels: low, medium, high, and extremely high.

[0013] Furthermore, in step S6, the graded early warning includes three levels: yellow, orange, and red, corresponding to medium, high, and extremely high-risk employees, respectively; personalized intervention measures include personalized training, on-site guidance, safety talks, and job adjustments.

[0014] Furthermore, in step S7, the model is validated quarterly, and an update is triggered when the model's accuracy is below 85% or its recall is below 80%; at the same time, automatic triggering conditions are set for changes in data distribution and business adjustments.

[0015] Furthermore, in step S6, the intervention task execution pool is initialized, parallel scheduling is performed according to priority, the task execution status is tracked in real time, intervention effect data is collected monthly, and the intervention strategy is dynamically adjusted.

[0016] This application also discloses an electronic device, including: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the digital-based coal mine job training method of the present invention.

[0017] This application also discloses a digital-based coal mine job training system, including: The multi-source data processing module is used to collect multi-source data on coal mine enterprise training, safety performance, equipment operation, management and employees, perform data cleaning, completion and standardization processing, and build a unified coal mine safety data warehouse with a star-shaped model architecture. The interference factor isolation module is used to classify and quantify four categories of non-training interference factors: equipment, management, environment, and individuals. It constructs an interference factor control matrix and eliminates multicollinearity. Through regression analysis, it calculates the net training effect for each employee. The causal association analysis module is used to construct a causal association model between training and safety performance using training variables as treatment variables and net training effect as outcome variables, and to quantify the causal effects of different training variables on safety performance. The risk feature management module is used to extract multi-dimensional safety risk features of employees, complete feature quantification, normalization, screening and dimensionality reduction processing, and establish and maintain an employee safety risk feature library. The safety risk prediction module is used to build a safety risk prediction model based on the employee safety risk feature library and historical accident data, calculate the employee safety risk index and classify the risk level; The risk warning and intervention module is used to generate graded warning information based on the employee's safety risk level, and to formulate and push personalized intervention measures based on the employee's risk characteristics. The model iteration and optimization module is used to periodically collect new data to update the coal mine safety data warehouse and the employee safety risk feature database, and to verify and iteratively optimize the causal relationship model and the safety risk prediction model.

[0018] Beneficial Effects: This application, through multi-source data integration and standardized processing, breaks down data silos between various business systems in coal mining enterprises, achieving centralized management and sharing of safety-related data. This provides a high-quality data foundation for subsequent analysis and modeling, solving the problems of data dispersion and inability to perform cross-dimensional fusion analysis under traditional models. By isolating non-training interference factors such as equipment, management, environment, and individuals, the application accurately separates the individual effects of training on safety performance. Then, using causal inference methods, it constructs a correlation model between training and safety performance, overcoming the limitations of traditional correlation analysis. This approach accurately quantifies the actual effects of different training elements and identifies group differences in training effectiveness, providing a scientific basis for the optimal allocation of training resources and significantly improving the relevance and effectiveness of training.

[0019] This application extracts and scientifically processes multi-dimensional safety risk characteristics of employees to construct a comprehensive employee safety risk characteristic database. Based on this database, a safety risk prediction model is established, shifting safety risk management from post-event handling to pre-event early warning. This enables early identification of high-risk employees and pinpointing their core risk points, significantly improving the accuracy of risk identification and the timeliness of early warnings. Simultaneously, a tiered early warning and precise intervention system is constructed, combining parallel task scheduling, real-time process monitoring, and quantitative effect evaluation to form a closed-loop management system from risk identification to intervention optimization, significantly improving the efficiency and effectiveness of intervention measures. Furthermore, the established automated model iteration mechanism adapts to the dynamic changes in the coal mine production environment and business operations, ensuring the long-term accuracy and adaptability of the model while reducing system maintenance costs. The full-process audit and traceability functions also meet the compliance requirements of coal mine safety management. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of a digital-based coal mine job training method according to the present invention; Figure 2 is a flowchart of multi-source data acquisition and standardization processing in an embodiment of the present invention; Figure 3 is a flowchart of the isolation and control of non-training interference factors in an embodiment of the present invention; Figure 4 is a flowchart of the construction of the causal relationship model between training and safety performance in an embodiment of the present invention; Figure 5 is a flowchart of the employee safety risk feature database construction process in an embodiment of the present invention; Figure 6 is a flowchart of the security risk prediction model construction in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention provides a digital-based method for coal mine job training, such as... Figure 1 As shown, the steps include: S1. Collect multi-source data on coal mine enterprise training, safety performance, equipment operation, management, and employee basics. After cleaning, completion, and standardization, establish a unified coal mine safety data warehouse with a star-shaped model architecture. S2. Classify and quantify four categories of non-training interference factors: equipment, management, environment, and individuals. Construct a control matrix for interference factors and eliminate multicollinearity. Calculate the net training effect for each employee through regression analysis. S3. Using training variables as treatment variables and net training effect as outcome variables, a causal association model is constructed using causal inference methods to quantify the causal effects of different training variables on safety performance; S4. Extract multi-dimensional safety risk characteristics of employees, and after quantification, normalization, screening and dimensionality reduction, establish an employee safety risk characteristic database; S5. Construct a safety risk prediction model based on the feature library and historical accident data, calculate the employee safety risk index and classify the risk level; S6. Generate graded early warning information based on risk levels, and formulate personalized intervention measures based on employee risk characteristics; S7. Regularly collect new data to update the data warehouse and feature library, and verify and iteratively optimize the causal relationship model and the security risk prediction model.

[0023] S1 represents multi-source data acquisition and standardization processing; such as Figure 2 As shown, this step is used to comprehensively collect multi-source heterogeneous data related to coal mining enterprises' safety training and safety performance. Through cleaning, deduplication, completion, and standardization, data noise and dimensional differences are eliminated to establish a unified coal mine safety data warehouse, providing high-quality data input for subsequent model construction.

[0024] Specifically, the data collection scope includes five categories: training data, safety performance data, equipment operation data, management data, and basic employee data. Training data, including employee training registration records, training attendance records, training duration records, training course content, training instructor information, training methods (online / offline, theory / practice), training assessment results, retake records, training satisfaction survey results, etc., are collected in real time from the digital training system. Safety performance data, including employee violation records, accident records, hazard investigation records, safety reward records, safety penalty records, and job operation assessment records, are collected in real time from the safety management system. Equipment operation data, including the operating time, fault records, maintenance records, update records, and equipment availability of various types of equipment in underground coal mines, are collected daily from the equipment management system. Management data, including safety management system update records, safety inspection records, safety meeting minutes, team management records, and on-site management records of coal mining enterprises, is collected weekly from the office automation system and safety management system. Employee basic data, including employee name, employee number, gender, age, length of service, position, education, major, skill level, work experience, and health status, is collected monthly from the human resources management system and updated in real time when information changes.

[0025] During the data cleaning process, duplicate data is deleted based on unique identifiers (employee number, equipment number, record number, etc.), and data with formatting and logical errors is corrected; missing numerical data is filled with the average value of the same position and the same time period, and missing categorical data is filled with the mode of the same position and the same time period; obviously abnormal data such as training time exceeding normal working hours and abnormal number of violation records are manually verified and corrected.

[0026] When standardizing data, the Z-score standardization method is used for numerical data. The calculation formula is as follows: ; in, The original data values, This is the average value of the data column. This represents the standard deviation of the data column.

[0027] One-hot coding is used for categorized data. For example, the training methods "online" and "offline" are coded as follows: and The data warehouse adopts a star schema design, with the employee dimension table at the center, and is linked to training fact tables, safety performance fact tables, equipment operation fact tables, and management fact tables. Role-based access control permissions are set to ensure data security.

[0028] It is clear that comprehensive collection and standardized processing of multi-source data are the foundation for all subsequent analysis and modeling, and data quality directly determines the accuracy and reliability of the model. Adopting a unified data warehouse can break down data silos between various business systems, enabling centralized data management and sharing, and providing possibilities for cross-dimensional data analysis.

[0029] This step enables the unified collection and standardized management of safety-related data in coal mining enterprises, providing reliable data support for subsequent isolation of interfering factors, construction of causal relationship models, and prediction of safety risks.

[0030] Among them, multi-source heterogeneous data refers to a data set that comes from different business systems and has different data formats and storage methods; Z-score standardization is a standardization method that converts raw data into a standardization method with a mean of 0 and a standard deviation of 1; one-hot coding is an encoding method that converts categorical features into multiple binary features; star schema is a data warehouse modeling method that takes the fact table as the center and connects multiple dimension tables.

[0031] Optionally, for security performance data with high real-time requirements, Kafka message queues can be used for streaming collection; for missing numerical data, in addition to filling with the average value of the same position in the same time period, linear interpolation or K-nearest neighbor filling can also be used; the data warehouse can be deployed on local servers or cloud platforms according to the size of the enterprise.

[0032] S2 is for isolating and controlling interfering factors; such as Figure 3 As shown, this step is used to classify and quantify non-training factors that affect coal mine safety performance, establish a control matrix of interference factors, isolate the impact of non-training factors on safety performance through statistical control methods, and obtain the change in safety performance caused only by training factors (net training effect).

[0033] Specifically, non-training factors are first categorized into four main types: equipment factors, management factors, environmental factors, and personal factors, and then quantified. Equipment factors include equipment aging (equipment service life / design service life), equipment failure rate (number of failures / operating time), equipment integrity rate (number of intact units / total number of units), and equipment replacement rate (number of new units / total number of units). Management factors include the frequency of safety inspections (number of inspections per month), the completeness of safety management systems (number of established systems / number of systems that should be established), on-site management level (on-site hazard rectification rate), and team management capability (average team safety performance). Environmental factors include underground gas concentration, dust concentration, temperature, humidity, and ventilation conditions, with daily monitoring averages used as quantified values. Personal factors include age, length of service, education level (1-5 corresponding to primary school to undergraduate and above), skill level (1-4 corresponding to junior worker to technician), and health status (1-3 corresponding to healthy to poor).

[0034] A control matrix for interference factors was constructed, with employees as rows and interference factors as columns. The correlation coefficients between different interference factors were calculated, and factors with absolute correlation coefficients greater than 0.8 were eliminated to remove multicollinearity. An expert group composed of coal mine safety management experts, technical experts, and frontline team leaders was invited to determine the weights of each interference factor using the analytic hierarchy process (AHP). The consistency ratio (CR) of the judgment matrix was determined to be less than 0.1. A multiple linear regression model was established with the comprehensive safety performance index as the dependent variable and the quantified values ​​of each interference factor as the independent variable. ; in, To form a comprehensive safety performance index, Quantify the values ​​of each interfering factor. For constant terms, For regression coefficients, This is the random error term.

[0035] The formula for calculating the comprehensive safety performance index is as follows: ; The value ranges from 0 to 100, with a higher index indicating better safety performance. The regression coefficients are estimated using the least squares method, and the model's coefficient of determination is required. The value is greater than 0.6. Finally, the expected safety performance value for each employee is calculated based on the regression model. The net training effect is obtained by subtracting the expected safety performance value from the actual safety performance value, i.e., net training effect = actual safety performance value - expected safety performance value. It is known that coal mine safety performance is the result of multiple factors working together. Without isolating the influence of non-training factors, the actual effectiveness of training cannot be accurately assessed. By establishing a control matrix of interference factors and a multiple linear regression model, the contribution of each non-training factor can be quantified, thus separating the individual effect of training factors and solving the core problem that existing technologies cannot distinguish the influence of training effects from other factors. Through this step, the influence of non-training factors such as equipment, management, environment, and individuals on safety performance is successfully isolated, enabling accurate calculation of the net training effect for each employee.

[0036] Among them, interfering factors refer to all factors other than training that may affect coal mine safety performance; multicollinearity refers to the phenomenon of a high linear correlation between multiple independent variables; the analytic hierarchy process (AHP) is a multi-criteria decision-making method that combines qualitative and quantitative analysis to determine the weight of each factor.

[0037] Optionally, for interference factors with strong nonlinear relationships, a multinomial regression model can be used instead of a multiple linear regression model; in addition to the analytic hierarchy process (AHP), the entropy weight method or the CRITIC method can be used to determine the weights; in addition to the paired t-test, the Wilcoxon signed-rank test can be used to test the significance of the net training effect.

[0038] S3 is used to construct a causal relationship model between training data and safety performance data; for example... Figure 4 As shown, this step is used to establish a causal relationship model between training data and safety performance data based on safety performance data after isolating interference factors and standardized training data, using a causal inference method, to calculate the causal effect value of different training variables on safety performance, and to quantify the actual impact of training on safety performance.

[0039] Specifically, the training content, training method, training duration, and training frequency are defined as treatment variables, and the net effect of training is defined as the outcome variable. The training content variable is divided into six categories: safety laws and regulations, safe operating procedures, accident case analysis, emergency rescue knowledge, equipment operation skills, and occupational health knowledge. The training method variable is divided into five categories: online theoretical training, offline theoretical training, on-site practical training, simulation training, and mentor-apprentice pairing training. The training duration variable is divided into three categories: short-term training (less than 1 day), medium-term training (1-3 days), and long-term training (more than 3 days). The training frequency variable is divided into four categories: once a month, once a quarter, once every six months, and once a year.

[0040] The propensity score was estimated using a logistic regression model, with covariates including employee age, length of service, education level, skill level, job position, and historical safety performance. The calculation formula is as follows: ;in, They indicated they would accept the training. This indicates that no training was received. As covariates, For constant terms, is the regression coefficient.

[0041] Nearest neighbor matching was used to match a control group to each trained employee. The matching caliper value was set to 0.05, and the standardized mean difference (SMD) of all covariates after matching had to be less than 0.1. The average treatment effect of different training variables on safety performance was calculated: ;in, For the net training effect of the employees receiving training, The net effect of training for employees who did not receive training.

[0042] The standard error and 95% confidence interval were calculated using the bootstrap method. If the confidence interval did not contain 0, the training variable was considered to have a significant causal effect on safety performance.

[0043] Employees were grouped by age, length of service, job position, and skill level. The average treatment effect was calculated for each group to analyze the heterogeneity of training effectiveness. The final causal relationship model was constructed as follows: ;in, For the first The weights of each training variable, For the first The average treatment effect of each training variable For the first The weight of individual employee groups For the first The training variable for the first... Heterogeneity treatment effect of individual employee groups.

[0044] It is known that traditional correlation analysis can only demonstrate a link between training and safety performance, but cannot prove a causal relationship. Employing causal inference methods such as propensity score matching can eliminate the influence of employee self-selection bias, accurately identify the causal effect of training on safety performance, and thus provide a scientific basis for the rational allocation of training resources. Through this step, a causal association model between training data and safety performance data was established. This model can accurately calculate the causal effect values ​​of different training content, methods, durations, and frequencies on safety performance, as well as the heterogeneity of training effectiveness among different employee groups, providing a quantitative basis for coal mining enterprises to optimize training programs.

[0045] Causal inference is a statistical method used to identify causal relationships rather than correlations between variables; propensity score matching (PSM) is a causal inference method used to eliminate selection bias by simulating the effects of randomized controlled trials by matching treatment groups with control groups that have similar characteristics; average treatment effect (ATE) refers to the average causal influence of the treatment variable on the outcome variable. Optionally, propensity score matching can use radius matching or kernel matching in addition to nearest neighbor matching; causal effect calculation can use difference-in-differences (DID) or instrumental variable methods in addition to propensity score matching; heterogeneity analysis can group variables by age, length of service, education level, or skill level in addition to age and length of service.

[0046] S4 is used to build an employee safety risk profile database; such as Figure 5 As shown, this step is used to extract multi-dimensional safety risk characteristics of employees, quantify, normalize, and select all characteristics to establish an employee safety risk characteristic library, providing input data for the safety risk prediction model. Specifically, four types of safety risk characteristics are extracted from standardized training data and safety performance data: Learning behavior characteristics include training attendance rate, training lateness rate, training early departure rate, online training learning time, online training video viewing completion rate, training interaction frequency, training question frequency, training note completion rate, etc. Exam result characteristics include: theoretical exam scores, practical exam scores, exam pass rate, number of retakes, records of cheating, and fluctuation range of exam scores (standard deviation of the scores of the three most recent exams). Historical violation characteristics: including total number of violations, number of general violations, number of serious violations, violation frequency (number of violations / working hours), time of the most recent violation, and distribution of violation types, etc. Job operation characteristics include job operation assessment scores, number of operational errors, number of hidden dangers identified, timeliness of hidden danger reporting, and performance in emergency drills.

[0047] Qualitative features are quantified. For example, the number of training interactions is divided into four levels: "very few (0-2 times)," "average (3-5 times)," "quite a few (6-10 times)," and "very many (more than 10 times)," assigned values ​​of 1, 2, 3, and 4 respectively. The min-max normalization method is used to unify the range of all feature values ​​to the [0,1] interval. The calculation formula is as follows: For inverse features (such as the number of violations, the number of operational errors, etc.), the following formula is used for normalization: Analysis of variance was used to screen features significantly associated with safety risks, and these features were retained. The characteristics of the data are analyzed; principal component analysis is used for dimensionality reduction, retaining principal components with a cumulative variance contribution rate greater than 85%. An employee safety risk feature database is established, with each record corresponding to one employee, including employee ID, name, position, and principal component feature values. It is stored in a relational database, indexed to improve query efficiency, and updated monthly, with real-time updates in case of significant data changes.

[0048] It is known that employee safety risks are determined by a variety of factors, and relying solely on exam scores as an evaluation indicator cannot fully reflect the actual safety risks of employees. By extracting multi-dimensional features and processing them scientifically, a comprehensive and accurate profile of employee safety risks can be constructed, providing high-quality input data for safety risk prediction models.

[0049] Feature extraction is the process of extracting features that reflect the essential attributes of things from raw data; min-max normalization is a normalization method that linearly transforms raw data to the [0,1] interval; principal component analysis (PCA) is an unsupervised dimensionality reduction method that transforms multiple related features into a few unrelated principal components through linear transformation. Optionally, in addition to analysis of variance, feature selection can also use chi-square test or mutual information method; in addition to principal component analysis, dimensionality reduction can also use linear discriminant analysis (LDA) or t-SNE; the feature library can be stored in a non-relational database (such as MongoDB) to improve the processing efficiency of unstructured data.

[0050] S5 is used for building security risk prediction models; such as Figure 6As shown, this step is used to construct a safety risk prediction model based on the employee safety risk feature database and historical safety accident data, using multivariate statistical analysis methods, to calculate the safety risk index of each employee and classify risk levels, and to identify high-risk employees.

[0051] Specifically, historical sample data from the past three years were extracted and divided into training and test sets in a 7:3 ratio. Employees were categorized into two groups: safe employees (no violations or accidents during the sample period) and high-risk employees (serious violations or minor accidents or above during the sample period). Employee category was used as the dependent variable, and principal component eigenvalues ​​were used as independent variables. A binary classification model was constructed using logistic regression, with the following formula: ;in, Indicates high-risk employees, Indicates safety staff, Principal component eigenvalues, For constant terms, is the regression coefficient.

[0052] Regression coefficients were estimated using maximum likelihood estimation, and L1 regularization was employed to avoid overfitting. The regularization parameter was determined through 5-fold cross-validation. The model was validated using a test set, and evaluation metrics included accuracy, precision, recall, and F1 score, requiring accuracy greater than 85% and recall greater than 80%. The safety risk index was obtained by multiplying the model's output probability values ​​by 100, and divided into four levels: low risk (safety risk index < 30); medium risk (30 ≤ safety risk index < 60); high risk (60 ≤ safety risk index < 80); and very high risk (safety risk index ≥ 80). The contribution of each principal component feature to the safety risk index was calculated to identify the main risk points for high-risk employees.

[0053] It is clear that the core objective of safety risk prediction is to identify as many truly high-risk employees as possible; therefore, recall rate is the most important evaluation metric. Logistic regression models offer advantages such as strong interpretability, high computational efficiency, and ease of deployment, making them highly suitable for practical applications in coal mining enterprises.

[0054] Logistic regression is a statistical model used for binary classification problems, calculating the probability that a sample belongs to a certain class. L1 regularization is a method to prevent overfitting by applying an L1 penalty to the regression coefficients, causing some coefficients to become zero, thus achieving feature selection. Recall refers to the proportion of samples that are actually positive but are correctly predicted as positive by the model, and is the most important evaluation indicator in safety risk prediction. Optionally, in addition to logistic regression, support vector machines (SVM) or random forests can be used for safety risk prediction models; in addition to L1 regularization, L2 regularization or elastic nets can be used for regularization methods; the threshold for risk level classification can be adjusted according to the actual situation of the enterprise.

[0055] S6 represents risk warning and precise intervention. This step generates tiered warning information and formulates targeted intervention measures based on the employee's safety risk level and characteristics. It provides timely intervention for high-risk employees to reduce their safety risks. Specifically, a tiered risk warning mechanism is established. Low-risk employees do not generate warning information, but receive regular routine safety training. Medium-risk employees receive a yellow warning, sent to the employee and their shift leader, reminding them to pay attention to safe operations and strengthen self-learning. High-risk employees receive an orange warning, sent to the employee, shift leader, team management personnel, and safety management department, requiring focused attention and targeted intervention. Extremely high-risk employees receive a red warning, sent to the employee, shift leader, team management personnel, safety management department, and mine-level leaders, immediately suspending the employee's underground work qualification, requiring mandatory safety training and assessment, and only allowing resumption of work after passing the assessment.

[0056] Personalized intervention measures are developed based on the risk characteristics of high-risk employees; for employees with weak safety knowledge, training on safety laws, regulations, and operating procedures is increased; for employees with poor operating skills, on-site practical training and simulation training are increased; for employees with habitual violations, accident case analysis and warning education are increased; high-risk employees who are not suitable for their current positions are transferred to auxiliary positions with lower safety risks; experienced senior employees or team leaders are assigned to provide one-on-one on-site guidance to correct unsafe behaviors, with a guidance period of 1-2 weeks; team managers or safety management personnel conduct safety talks with high-risk employees to understand their thoughts and work difficulties and improve their safety awareness; high-risk employees are subject to regular safety assessments, and those who fail the assessments continue to receive training and guidance.

[0057] The effectiveness of interventions is assessed monthly, with evaluation indicators including changes in the safety risk index, changes in the number of violations, and changes in the number of operational errors. An intervention effectiveness file is established to record the intervention process and its results. It is known that the causes of safety risks differ among employees, and a uniform intervention approach has limited effectiveness. Through tiered early warning and precise intervention, personalized plans can be developed based on the specific risk characteristics of employees, significantly improving the effectiveness of interventions and reducing the probability of safety accidents. This step establishes a comprehensive risk early warning and precise intervention mechanism, enabling differentiated early warning and intervention measures for employees with different risk levels. The intervention effectiveness rate for high-risk employees reaches over 85%, significantly reducing the accident rate in coal mining enterprises.

[0058] Among these, tiered early warning is a mechanism for taking early warning measures of varying intensities based on risk levels; precise intervention is the process of developing personalized intervention plans based on the specific risk characteristics of employees; and intervention effectiveness evaluation is the process of quantitatively assessing the effectiveness of intervention measures. Optionally, early warning information can be released through mobile office platforms such as WeChat and DingTalk; intervention measures, in addition to personalized training and job adjustments, can also include psychological counseling and safety point rewards; the intervention effectiveness evaluation cycle can be adjusted to 2 weeks or 1 month depending on the company's actual situation.

[0059] S7 is for model iteration and optimization. This step involves periodically collecting new training and safety performance data to validate and update the causal correlation model and safety risk prediction model, continuously optimizing model parameters and improving model accuracy and reliability. Specifically, new training and safety performance data are collected monthly, preprocessed and standardized according to the method in S1, and the coal mine safety data warehouse and employee safety risk feature database are updated. The causal correlation model and safety risk prediction model are validated quarterly. For the causal correlation model, the average treatment effect of each training variable is calculated using new data and compared with the original model parameters. If the difference exceeds 10%, the model is updated. For the safety risk prediction model, accuracy, precision, recall, and F1 score are calculated using a new test set. If the accuracy is below 85% or the recall is below 80%, the model is updated.

[0060] When the model needs updating, it is rebuilt using all the latest data to ensure that the new model outperforms the original. Based on the actual situation and feedback from coal mining enterprises, the model structure and parameters are continuously optimized, such as introducing new training variables and equipment factors, adjusting the weights of interference factors, and optimizing the risk level classification thresholds.

[0061] It is known that the production environment, safety management system, and employee structure of coal mining enterprises are constantly changing, and the performance of the model will gradually decline over time. By establishing a sound model iteration and optimization mechanism, it is possible to ensure that the model adapts to constantly changing application scenarios and maintains high accuracy and reliability in the long term.

[0062] This section fully elaborates on the core basic process of digital coal mine job training methods. Through multi-source data integration, isolation of interfering factors, causal correlation analysis, and safety risk prediction, it solves the three core problems of existing technologies: inability to quantify training effectiveness, inability to distinguish the impact of training on other factors, and inability to identify high-risk employees in advance.

[0063] Model validation is the process of evaluating model performance using new data; model update is the process of retraining the model using the latest data; and model optimization is the process of adjusting the model structure and parameters based on feedback from real-world applications. Optionally, the model validation cycle can be adjusted to once a month or once a quarter, depending on the company's data update frequency; model updates can use incremental learning methods, eliminating the need to retrain all data; and a user feedback mechanism can be introduced to optimize the model based on the opinions of managers and employees.

[0064] Based on the core methodology and basic process described above, we further refine the parallel execution process of intervention tasks, the real-time monitoring mechanism, and the closed-loop management of effects to improve the execution efficiency and effectiveness of intervention measures. At the same time, we improve the automated iteration process of the model and reduce system maintenance costs.

[0065] Furthermore, this includes the initialization and parallel scheduling of the intervention task pool. This step initializes the intervention task execution pool and performs parallel scheduling based on system resources and intervention task priorities to improve the execution efficiency of intervention tasks. Specifically, the maximum number of concurrent tasks in the intervention task execution pool is configured through the `max_jobs` parameter, with a default value of twice the number of system CPU cores. Intervention tasks are divided into three priorities: extremely high priority (mandatory safety training and assessment tasks corresponding to red alerts); high priority (personalized training and on-site guidance tasks corresponding to orange alerts); and medium priority (routine reminders and self-learning tasks corresponding to yellow alerts). The task execution pool is a technology used to manage concurrent task execution, limiting the number of tasks executed simultaneously to avoid exhausting system resources. Task priorities are assigned based on the urgency and importance of tasks, with higher-priority tasks executed first. Optionally, the task execution pool can be implemented using a thread pool or a process pool. In addition to priority scheduling, task scheduling can also use time-slice round-robin scheduling or shortest job first scheduling. The size of the task execution pool can be dynamically adjusted according to system load.

[0066] A priority scheduling algorithm is employed, allocating system resources to high-priority tasks first. When system resources are insufficient, the execution of low-priority tasks is paused and resumed once resources are released. Given the large number of employees and intervention tasks generated simultaneously in coal mining enterprises, a serial execution approach would result in significant task delays. By establishing an intervention task execution pool and employing parallel scheduling, system resources can be fully utilized, the execution efficiency of intervention tasks can be improved, and timely intervention for high-risk employees can be ensured.

[0067] Furthermore, this includes the distribution and execution of intervention tasks. This step distributes the generated intervention tasks to the corresponding personnel and systems, tracks the execution status of the tasks, and ensures that the intervention measures are effectively implemented. Specifically, based on the type of intervention task, tasks are distributed to the corresponding personnel and systems. Personalized training tasks are distributed to the digital training system, automatically assigning training courses and assessments to employees; on-site coaching tasks are distributed to the corresponding team leaders or senior employees, generating on-site coaching plans; safety conversation tasks are distributed to the corresponding team managers or safety management personnel, generating conversation record templates; and job adjustment tasks are distributed to the human resources management system, generating job adjustment approval processes.

[0068] The system tracks the execution status of each task in real time, including whether it has not started, is in progress, has been completed, or has been canceled, and generates timeout reminders for tasks that are not completed within the specified time. It is clear that the effectiveness of intervention measures depends on the quality and timeliness of their execution. Through a standardized task distribution and execution tracking mechanism, it is possible to ensure that each intervention task is effectively implemented, avoiding situations where there are only warnings but no interventions.

[0069] Task distribution is the process of assigning tasks to specific personnel or systems; task status tracking is the process of monitoring task execution progress and status in real time. Optionally, task distribution can be implemented through a workflow engine; it can support manual adjustment and reassignment of tasks; and it can generate task execution reminders to avoid task omissions.

[0070] Furthermore, this includes real-time monitoring of the intervention process. This step is used to monitor various indicators during the intervention process in real time, promptly identify and address any anomalies, and ensure the smooth progress of the intervention. Specifically, the core indicators monitored include: Training-related indicators include training attendance rate, training completion rate, and assessment pass rate; on-site indicators include the number of on-site coaching sessions, coaching duration, and number of times unsafe behaviors were corrected; management indicators include the number of safety talks, job adjustment completion rate, and intervention task completion rate.

[0071] When monitored indicators fall below preset thresholds, an anomaly alarm is automatically generated and sent to relevant management personnel. Management personnel then take timely action based on the alarm information, such as reminding employees to attend training or urging team leaders to provide on-site guidance. It is understood that various anomalies may occur during the intervention process; failure to detect and address them promptly will severely impact the intervention's effectiveness. Real-time monitoring and anomaly alarm mechanisms enable the timely identification and resolution of problems during intervention, ensuring the effectiveness of intervention measures.

[0072] Abnormal situations refer to circumstances that may affect the effectiveness of the intervention, such as employees not attending training or on-site guidance not being conducted as planned. Optionally, a large visual monitoring screen can be used to display the overall situation of the intervention process; abnormal thresholds can be set, and alarms can be automatically generated when indicators exceed the thresholds; manual reporting and handling of abnormal situations can be supported.

[0073] Furthermore, this includes the collection and analysis of intervention effect data. This step is used to collect effect data after the implementation of intervention measures, to quantitatively analyze the intervention effect, and to evaluate the effectiveness of the intervention measures. Specifically, the intervention effect evaluation cycle is one month, and the collected effect data includes: changes in employees' safety risk index; changes in the number of employee violations, general hazards, and serious hazards; changes in the number of employee operational errors and job operation assessment scores; and changes in employee training attendance rate, training completion rate, and assessment scores.

[0074] Statistical analysis of the collected data is performed to calculate the effectiveness rate of each intervention. The effects of different interventions on employees with different risk profiles are compared to select the optimal intervention plan. It is clear that intervention effectiveness evaluation is a crucial step in closed-loop management. Quantitative evaluation allows us to understand the effectiveness of interventions and provides a basis for optimizing subsequent intervention plans. This step achieves a quantitative evaluation of intervention effectiveness, accurately identifying effective interventions and providing data support for subsequent precise interventions.

[0075] Among them, effectiveness data refers to data that reflects the effectiveness of intervention measures, such as changes in employees' safety risk index, changes in the number of violations, and changes in the number of operational errors. Optionally, effectiveness data can be automatically collected from various business systems without manual input; personalized intervention effectiveness analysis reports can be generated; and the effects of different intervention measures can be compared to select the optimal intervention plan.

[0076] Furthermore, this includes dynamic adjustment of intervention strategies. This step is used to dynamically adjust intervention strategies, optimize intervention measures, and improve intervention effectiveness based on the intervention effectiveness evaluation results. Specifically, based on the intervention effectiveness evaluation results, the intervention strategies are adjusted as follows: for employees with effective intervention, the intervention intensity is reduced and converted to routine monitoring; for employees with insignificant intervention effects, the reasons are analyzed and intervention measures are adjusted, such as increasing training duration, changing training methods, and extending the on-site coaching cycle; for employees with ineffective intervention, intervention measures are upgraded, such as conducting mandatory safety training, adjusting job positions, and reporting to mine-level leadership.

[0077] Effective intervention strategies are stored in an intervention strategy knowledge base to provide a reference for subsequent interventions on employees with similar risk characteristics. It is known that different employees respond differently to interventions, and fixed intervention strategies cannot adapt to all situations. By dynamically adjusting intervention strategies, interventions can be optimized based on actual employee responses, significantly improving intervention effectiveness. Through this step, the targeting and effectiveness of intervention strategies are significantly enhanced.

[0078] Dynamic adjustment of intervention strategies involves adjusting the type, intensity, and duration of intervention measures in real time based on feedback on intervention effectiveness. Optionally, reinforcement learning algorithms can be used to automatically optimize intervention strategies; intervention measures can be adjusted based on employee feedback; and an intervention strategy knowledge base can be established to accumulate best practices.

[0079] Furthermore, this includes automated model iteration triggering. This step is used to automatically trigger the model verification and update process based on preset trigger conditions, thereby automating model iteration and reducing system maintenance costs. Specifically, the following automated model iteration trigger conditions are set: periodic triggering, i.e., automatically triggering model verification once per quarter; performance triggering, i.e., automatically triggering model updates when the model's accuracy is below 85% or recall is below 80%; data triggering, i.e., automatically triggering model updates when the difference between the distribution of new data and the distribution of original data exceeds a preset threshold; and business triggering, i.e., automatically triggering model updates when coal mining enterprises introduce new training methods, new equipment, or undergo significant adjustments to safety management systems.

[0080] When the triggering conditions are met, the system automatically executes the model verification and update process, generates a model iteration report, and notifies relevant management personnel. It is known that manual model iteration requires professional data analysts, which is costly and inefficient. By setting automated triggering conditions, model iteration can be automated, reducing system maintenance costs and ensuring the timeliness of the model. Through this step, the manual workload of model iteration is reduced by more than 80%, and the timeliness of model updates is significantly improved, enabling rapid adaptation to changes in the business of coal mining enterprises.

[0081] Triggering conditions refer to events or indicators that can trigger model iteration, such as a decline in model performance, a significant change in data distribution, or a major adjustment to the company's management system. Optionally, multiple triggering conditions can be set, and model iteration can be triggered when any one condition is met; manual triggering of model iteration is supported; and model iteration reports can be generated to record changes in the model.

[0082] Furthermore, this includes full-process auditing and traceability. This step is used to record all operations and data throughout the training and risk management process, enabling full-process auditing and traceability to meet the safety management compliance requirements of coal mining enterprises. Specifically, the recorded audit data includes: all operation records of data collection and processing; all parameters and results of model training, validation, and updates; records of risk warning generation and issuance; records of intervention task distribution, execution, and effects; and all user login and operation records.

[0083] Audit data must be retained for at least three years and can be queried and exported by time, personnel, and event, generating audit reports that meet coal mine safety regulatory requirements. It is well known that coal mine safety production has strict compliance requirements, and full-process auditing and traceability are necessary conditions for meeting regulatory requirements, while also providing a basis for accident investigation and liability determination. This step achieves auditability and traceability of the entire training and risk management process, meeting the safety management compliance requirements of coal mining enterprises and providing a reliable basis for accident investigation and liability determination.

[0084] The full-process audit is a process of reviewing and evaluating the entire training and risk management process; traceability refers to the ability to track the source and change history of each operation and data. Optionally, blockchain technology can be used to store audit data to ensure data immutability; standardized audit reports can be generated to meet regulatory requirements; and traceability queries can be supported by time, personnel, events, and other dimensions.

[0085] The present invention also provides a system for implementing the above-mentioned digital-based coal mine job training method, comprising: The multi-source data processing module is used to collect multi-source data on coal mine enterprise training, safety performance, equipment operation, management and employees, perform data cleaning, completion and standardization processing, and build a unified coal mine safety data warehouse with a star-shaped model architecture. The interference factor isolation module is used to classify and quantify four categories of non-training interference factors: equipment, management, environment, and individuals. It constructs an interference factor control matrix and eliminates multicollinearity. Through regression analysis, it calculates the net training effect for each employee. The causal association analysis module is used to construct a causal association model between training and safety performance using training variables as treatment variables and net training effect as outcome variables, and to quantify the causal effects of different training variables on safety performance. The risk feature management module is used to extract multi-dimensional safety risk features of employees, complete feature quantification, normalization, screening and dimensionality reduction processing, and establish and maintain an employee safety risk feature library. The safety risk prediction module is used to build a safety risk prediction model based on the employee safety risk feature library and historical accident data, calculate the employee safety risk index and classify the risk level; The risk warning and intervention module is used to generate graded warning information based on the employee's safety risk level, and to formulate and push personalized intervention measures based on the employee's risk characteristics. The model iteration and optimization module is used to periodically collect new data to update the coal mine safety data warehouse and the employee safety risk feature database, and to verify and iteratively optimize the causal relationship model and the safety risk prediction model.

[0086] This invention also provides an electronic device for implementing the above-mentioned digital-based coal mine job training method, and a computer-readable storage medium storing a computer program, used to transform the technical solution of this invention into an executable software product. An electronic device includes a processor, a memory, a bus, and a communication interface, wherein the processor, memory, and communication interface are connected via the bus and communicate with each other. The memory stores a computer program, which includes program instructions; the processor calls the program instructions in the memory to execute the steps of the above-mentioned digital-based coal mine job training method. Specifically, the electronic device further includes an input unit and an output unit. The input unit receives parameters and instructions input by the user, such as model parameter settings and intervention strategy adjustments; the output unit outputs the model's analysis results, risk warning information, intervention effect reports, etc., supporting visualization. The electronic device can be various computing devices such as desktop computers, laptops, servers, and mobile terminals, and can be deployed according to the actual needs of coal mining enterprises.

[0087] A computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the aforementioned digital-based coal mine job training method. The computer-readable storage medium can be any medium capable of storing a computer program, such as a read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0088] The scope of protection of this invention is not limited to the methods, apparatus, and electronic devices described above, but also includes any products, systems, and services that incorporate the core technical features of this invention. Any modifications, substitutions, and improvements made to this invention without departing from its spirit and principles should fall within the scope of protection of this invention.

[0089] This section transforms the technical solution of this invention into an executable software product, providing a complete implementation scheme for electronic devices and computer-readable storage media, facilitating deployment and application by coal mining enterprises. Adopting this technical solution enables rapid implementation of the method of this invention, providing strong technical support for safety training and risk management in coal mining enterprises.

[0090] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital-based coal mine job training method, characterized in that, Includes the following steps: S1. Collect multi-source data on coal mine enterprise training, safety performance, equipment operation, management, and employee basics. After cleaning, completion, and standardization, establish a unified coal mine safety data warehouse with a star-shaped model architecture. S2. Classify and quantify four categories of non-training interference factors: equipment, management, environment, and individuals. Construct a control matrix for interference factors and eliminate multicollinearity. Calculate the net training effect for each employee through regression analysis. S3. Using training variables as treatment variables and net training effect as outcome variables, a causal association model is constructed using causal inference methods to quantify the causal effects of different training variables on safety performance; S4. Extract multi-dimensional safety risk characteristics of employees, and after quantification, normalization, screening and dimensionality reduction, establish an employee safety risk characteristic database; S5. Construct a safety risk prediction model based on the feature library and historical accident data, calculate the employee safety risk index and classify the risk level; S6. Generate graded early warning information based on risk levels, and formulate personalized intervention measures based on employee risk characteristics; S7. Regularly collect new data to update the data warehouse and feature library, and verify and iteratively optimize the causal relationship model and the security risk prediction model.

2. The digital-based coal mine job training method according to claim 1, characterized in that, In step S1, Z-score standardization is used for numerical data and one-hot encoding is used for categorical data; the data collection frequency is divided into real-time, daily, weekly and monthly according to type, and the data integrity is not less than 98%.

3. The digital-based coal mine job training method according to claim 1, characterized in that, In step S2, the weights of each interfering factor are determined using the analytic hierarchy process (AHP). A regression model is established with the comprehensive safety performance index as the dependent variable. The net training effect is obtained by subtracting the expected safety performance value from the actual safety performance value.

4. The digital-based coal mine job training method according to claim 1, characterized in that, In step S3, the training variables include training content, training method, training duration, and training frequency; logistic regression is used to estimate propensity score, and nearest neighbor matching is used to match the treatment group with the control group to calculate the average treatment effect and the heterogeneous treatment effect.

5. The digital-based coal mine job training method according to claim 1, characterized in that, In step S4, the extracted safety risk features include four categories: learning behavior, exam results, historical violations, and job operations; the minimum-maximum normalization is used to unify the feature value range, and dimensionality reduction is performed through principal component analysis.

6. The digital-based coal mine job training method according to claim 1, characterized in that, In step S5, the historical samples are divided into training and test sets in a 7:3 ratio, and a prediction model is constructed using logistic regression with L1 regularization; the safety risk level is divided into four levels: low, medium, high, and extremely high.

7. The digital-based coal mine job training method according to claim 1, characterized in that, In step S6, the graded early warning includes three levels: yellow, orange, and red, corresponding to medium, high, and extremely high-risk employees, respectively; personalized intervention measures include personalized training, on-site guidance, safety talks, and job adjustments.

8. The digital-based coal mine job training method according to claim 1, characterized in that, In step S7, the model is validated quarterly, and an update is triggered when the model accuracy is below 85% or the recall is below 80%. At the same time, automatic triggering conditions are set for changes in data distribution and business adjustments.

9. A digital-based coal mine job training method according to claim 1, characterized in that, In step S6, the intervention task execution pool is initialized, and parallel scheduling is performed according to priority. The task execution status is tracked in real time, and intervention effect data is collected monthly and the intervention strategy is dynamically adjusted.

10. A digital-based coal mine job training method according to claim 1, characterized in that, include: The multi-source data processing module is used to collect multi-source data on coal mine enterprise training, safety performance, equipment operation, management and employees, perform data cleaning, completion and standardization processing, and build a unified coal mine safety data warehouse with a star-shaped model architecture. The interference factor isolation module is used to classify and quantify four categories of non-training interference factors: equipment, management, environment, and individuals. It constructs an interference factor control matrix and eliminates multicollinearity. Through regression analysis, it calculates the net training effect for each employee. The causal association analysis module is used to construct a causal association model between training and safety performance using training variables as treatment variables and net training effect as outcome variables, and to quantify the causal effects of different training variables on safety performance. The risk feature management module is used to extract multi-dimensional safety risk features of employees, complete feature quantification, normalization, screening and dimensionality reduction processing, and establish and maintain an employee safety risk feature library. The safety risk prediction module is used to build a safety risk prediction model based on the employee safety risk feature library and historical accident data, calculate the employee safety risk index and classify the risk level; The risk warning and intervention module is used to generate graded warning information based on the employee's safety risk level, and to formulate and push personalized intervention measures based on the employee's risk characteristics. The model iteration and optimization module is used to periodically collect new data to update the coal mine safety data warehouse and the employee safety risk feature database, and to verify and iteratively optimize the causal relationship model and the safety risk prediction model.

Images