A work injury prevention training system

Abstract

The present application relates to the technical field of data processing, in particular to a kind of work injury prevention training system, comprising: construction module, for obtaining the environmental data stream on real work site, and according to environmental data stream, the digital twin model corresponding to real work site is constructed;Simulation module, for carrying out work injury prevention training to user by digital twin model, and in the process of work injury prevention training, the behavior data of user is collected;Analysis module, for carrying out work injury analysis according to behavior data, generates work injury early warning information;Output module, for generating security credit points according to work injury early warning information and historical training records, so that work injury prevention training system can carry out accurate risk assessment and real-time early warning according to the specific behavior in the actual operation of user personal.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to a workplace injury prevention training system. Background Technology

[0002] Work injury prevention training is an important measure to protect workers' safety and health and reduce the occurrence of production accidents. Its effective implementation plays a key role in improving the inherent safety level of enterprises.

[0003] Existing workplace injury prevention training systems mostly employ offline centralized lectures or online learning platforms, conveying general safety regulations and operational knowledge to employees through methods such as playing safety education videos and answering safety knowledge quizzes. However, because the training content and assessment methods heavily rely on pre-set fixed teaching materials and exams, they cannot conduct dynamic and personalized risk assessments and feedback based on individual employees' real-time operational behaviors. This results in insufficient training relevance and makes it difficult to accurately identify and warn of specific safety risks for particular employees in actual work environments. Summary of the Invention

[0004] To address the technical problem that existing work injury prevention training systems cannot achieve targeted risk assessment and early warning, this application provides a work injury prevention training system.

[0005] The work injury prevention training system provided in this application adopts the following technical solution:

[0006] A workplace injury prevention training system includes:

[0007] The module is used to acquire environmental data streams from the real work site and build a digital twin model corresponding to the real work site based on the environmental data streams.

[0008] The simulation module is used to conduct workplace injury prevention training for users through a digital twin model, and to collect user behavior data during the workplace injury prevention training process;

[0009] The analysis module is used to perform work injury analysis based on behavioral data and generate work injury early warning information;

[0010] The output module is used to generate safety credit scores based on work injury early warning information and historical training records.

[0011] Furthermore, the steps of acquiring environmental data streams from the actual workplace and constructing a digital twin model corresponding to the actual workplace based on the environmental data streams include:

[0012] By deploying each sensor network point on a real work site, environmental data streams are acquired at each sensor network point.

[0013] Based on the edge computing nodes embedded in each sensor network point, each environmental data stream is processed to generate multiple environmental risk entropies.

[0014] Based on multiple environmental risk entropies, an environmental risk topology map of the actual workplace is generated.

[0015] Based on the entropy of each environmental risk in the environmental risk topology map, a grid resolution division threshold is set to re-divide the preset three-dimensional network of the real work site to obtain a multi-resolution three-dimensional grid.

[0016] Determine the environmental risk entropy of each grid cell in the multi-resolution 3D mesh. Load a preset texture-physical behavior model onto grid cells with environmental risk entropy higher than the preset environmental risk entropy, and load a preset texture-simplified physical model onto grid cells with environmental risk entropy lower than the preset environmental risk entropy to obtain a digital twin model.

[0017] Furthermore, the steps of processing each environmental data stream based on the edge computing nodes embedded in each sensor network point to generate multiple environmental risk entropies include:

[0018] Each edge computing node performs normalization processing on the corresponding environmental data stream to generate multiple environmental parameter vectors.

[0019] Calculate the fluctuation amplitude of each environmental parameter vector in each environmental dimension within the sliding time window to obtain multiple environmental fluctuation characteristics;

[0020] Based on the characteristics of environmental fluctuations, the uncertainty measures of each edge computing node in each environmental dimension are calculated, and then the uncertainty measures in each environmental dimension are fused to generate the environmental risk entropy corresponding to each edge computing node.

[0021] Furthermore, the steps of conducting workplace injury prevention training for users using digital twin models, and collecting user behavioral data during the training process, include:

[0022] Presenting a virtual scene of the venue to users through a digital twin model;

[0023] Based on the user's real-time operation in the virtual scene of the site, simulation is performed through a digital twin model to obtain the changes in environmental state and equipment interaction feedback caused by the real-time operation.

[0024] Collect user decision-making actions in response to changes in environmental conditions and device interaction feedback;

[0025] Quantify decision-making actions to obtain behavioral data.

[0026] Furthermore, the steps for generating work injury early warning information based on behavioral data include:

[0027] Feature extraction is performed on behavioral data to obtain behavioral feature vectors;

[0028] The safety deviation is obtained by calculating the behavioral feature vector through a pre-set safety assessment model.

[0029] The safety deviation is input into the preset risk assessment rules for risk matching to obtain the risk type and work injury risk level;

[0030] Work injury early warning information is generated based on risk type and work injury risk level.

[0031] Furthermore, the steps for generating safety credit scores based on work injury early warning information and historical training records include:

[0032] Work injury early warning information and historical training records are quantified to obtain safety assessment data;

[0033] Based on safety assessment data, work injury early warning information and historical training records are weighted to obtain a weighted assessment value;

[0034] Based on the weighted evaluation value, the user's score in their respective user group is calculated to obtain a relative safety score;

[0035] The relative safety score is mapped to a preset score range to obtain a safety credit score.

[0036] Furthermore, work injury early warning information includes the number of warnings and the warning level, and historical training records include training results and training completion rate. The steps for quantifying work injury early warning information and historical training records to obtain safety assessment data include:

[0037] By pre-setting a quantitative indicator system, the number of warnings, warning levels, training results, and training completion are processed with unified dimensions to obtain a set of quantitative indicators.

[0038] The quantitative indicators in the quantitative indicator set are weighted and integrated to generate a comprehensive security assessment value;

[0039] The comprehensive security assessment value is mapped to a preset security assessment range to obtain security assessment data.

[0040] Beneficial effects achieved:

[0041] This application provides a workplace injury prevention training system, comprising: a construction module for acquiring environmental data streams from a real workplace and constructing a digital twin model corresponding to the real workplace based on the environmental data streams; a simulation module for conducting workplace injury prevention training on users through the digital twin model and collecting user behavior data during the training process; an analysis module for performing workplace injury analysis based on the behavior data and generating workplace injury early warning information; and an output module for generating safety credit scores based on workplace injury early warning information and historical training records.

[0042] In this application, a highly realistic virtual training environment is created by constructing a digital twin model of the real workplace environment through a construction module. This ensures that the training scenario closely corresponds to real working conditions. Then, a simulation module provides workplace injury prevention training to users within the digital twin model and collects user behavior data in real time. This behavioral data reflects the user's specific operations and reactions in the virtual training environment, providing a foundation for personalized analysis. The analysis module then performs workplace injury analysis based on the behavioral data, generating workplace injury warning information. This analysis process directly targets the user's real-time operational behavior, achieving a shift from general assessment to individualized risk assessment. Finally, the output module generates a safety credit score based on the workplace injury warning information and historical training records. This safety credit score integrates current workplace injury warnings with historical performance, forming a dynamic and targeted feedback mechanism. This enables the workplace injury prevention training system to conduct accurate risk assessments and provide real-time warnings based on the user's specific actions during actual operations. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of a work injury prevention training system according to this application;

[0044] Figure 2 A flowchart illustrating the steps involved in constructing a digital twin model for this application;

[0045] Figure 3 This is a flowchart illustrating the steps involved in this application: conducting simulation training using a digital twin model and collecting user behavior data.

[0046] Figure 4 This is a flowchart illustrating the steps involved in generating work injury early warning information based on behavioral data in this application.

[0047] Figure 5 A flowchart illustrating the steps involved in generating a security credit score for this application.

[0048] Explanation of icon numbers:

[0049] 10. Construction Module; 20. Simulation Module; 30. Analysis Module; 40. Output Module. Detailed Implementation

[0050] The following is in conjunction with the appendix Figures 1-5 This application will be described in further detail.

[0051] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0052] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0053] This application discloses a workplace injury prevention training system.

[0054] Please refer to Figure 1 The work injury prevention training system proposed in this embodiment includes:

[0055] The module 10 is used to acquire environmental data streams from the actual workplace and construct a digital twin model corresponding to the actual workplace based on the environmental data streams; the simulation module 20 is used to conduct work injury prevention training for users through the digital twin model and collect user behavior data during the work injury prevention training process; the analysis module 30 is used to perform work injury analysis based on the behavior data and generate work injury early warning information; the output module 40 is used to generate safety credit scores based on work injury early warning information and historical training records.

[0056] The work injury prevention training system proposed in this embodiment establishes a dynamic assessment system that can realistically reflect individual operational differences. Module 10 transforms the real workplace into a digital twin model, providing a highly realistic interactive foundation for work injury prevention training. The virtual environment used for work injury prevention assessment is consistent with the actual work conditions of the real workplace. Simulation module 20 conducts work injury prevention training for users on this digital twin model and simultaneously records the behavioral data of the participating users. By capturing the personalized action habits and judgment logic exhibited by users in simulated operations, it provides a data foundation for subsequent individualized training analysis. Analysis module 30 generates work injury early warning information by parsing the behavioral data, realizing the transformation of abstract safety regulations into a quantitative evaluation of each user's specific operational behavior, thereby focusing risk assessment on the individual's actual operational level. Output module 40 integrates early warning information and historical records to generate safety credit scores, forming traceable and quantifiable personal safety performance indicators. Ultimately, it constructs a data-driven, dynamically adjustable personalized work injury prevention mechanism that effectively improves the accuracy of training assessment and the targeting of early warnings.

[0057] In one feasible implementation, refer to Figure 2 As shown, the specific execution steps of the construction module include steps S11 to S15:

[0058] Step S11: Obtain the environmental data stream at each sensor network point deployed on the actual work site.

[0059] By systematically deploying a sensor network consisting of multiple types of environmental sensors in key areas of the real work site, each sensor network point can continuously collect environmental parameters such as temperature, humidity, light intensity, and gas concentration at its location, forming a continuous environmental data stream. This deployment method ensures the spatial integrity and temporal continuity of the environmental data stream collection, providing a comprehensive and real-time data foundation for subsequent accurate risk assessment. Its most important effect is to build a continuous and reliable data association between the digital twin model and the real work site, enabling the construction of the virtual environment and risk simulation to be based on real and dynamic on-site environmental data.

[0060] Step S12: Based on the edge computing nodes embedded in each sensor network point, process each environmental data stream to generate multiple environmental risk entropies.

[0061] Based on edge computing nodes embedded in each sensor network point, the environmental data streams collected from the corresponding sensor network points are processed to generate environmental risk entropy corresponding to each sensor network point. This enables real-time localized calculation of environmental risks at the source of data collection, thereby avoiding the direct transmission of large amounts of environmental data streams to the central system, reducing network bandwidth pressure and improving response speed.

[0062] In this step, environmental data streams such as temperature, humidity, light intensity, and gas concentration are analyzed in real time through edge computing nodes. Multi-dimensional environmental parameters are fused and transformed into a single environmental risk entropy. This environmental risk entropy can quantitatively characterize the local environmental instability and risk intensity of the area where each sensor network point is located. It provides accurate and spatially distributed risk input data for generating an overall environmental risk topology map in subsequent steps, thereby ensuring that the foundation of the digital twin model construction is highly consistent with the dynamic risk status of the real work site.

[0063] Step S13: Generate an environmental risk topology map of the actual work site based on multiple environmental risk entropies.

[0064] By using spatial interpolation algorithms, the environmental risk entropy calculated at each sensor network point is used as a known data point. Based on its corresponding spatial coordinates, the estimated environmental risk entropy value of the area on the real work site where no sensor network points are deployed is calculated. This yields the environmental risk status of each point within the continuous spatial range of the entire real work site. This generates a complete two-dimensional or three-dimensional environmental risk topology map covering the real work site, with different colors or numerical gradients representing the risk level. This transforms the local risk information (i.e., environmental risk entropy) that was originally isolated at each sensor network point into a holistic, continuous, and visualized spatial risk distribution representation. This intuitively reveals the overall risk profile of the entire real work site and the spatial clustering of high-risk areas, providing a direct and accurate scientific basis for subsequent differentiated multi-resolution grid division of the digital twin model.

[0065] Step S14: Based on the entropy of each environmental risk in the environmental risk topology map, set the grid resolution division threshold to re-divide the preset three-dimensional network of the real work site and obtain a multi-resolution three-dimensional grid.

[0066] By statistical analysis, the distribution characteristics of each environmental risk entropy in the environmental risk topology map are determined. For example, the environmental risk entropy is divided into different intervals according to percentiles and corresponding resolution levels are set, thereby establishing a mapping rule between environmental risk entropy and grid fineness.

[0067] When re-dividing the preset 3D mesh, each mesh cell in the preset 3D mesh is first traversed, and its subdivision degree is dynamically adjusted according to the interval in which its corresponding environmental risk entropy falls. For high-risk areas where the environmental risk entropy is higher than the preset environmental risk entropy, the mesh is recursively subdivided to improve the resolution, while for low-risk areas where the environmental risk entropy is lower than the preset environmental risk entropy, the mesh is maintained or merged to reduce the resolution. Finally, a multi-resolution 3D mesh with varying density and matching risk distribution is formed, which realizes the optimized allocation of computing resources. This enables the digital twin model to effectively control the geometric complexity of the overall model while ensuring high-precision simulation of high-risk areas, and effectively improves the rendering efficiency and real-time interactivity of the subsequent training simulation process.

[0068] Step S15: Determine the environmental risk entropy of each grid cell in the multi-resolution 3D mesh. Load a preset texture-physical behavior model onto grid cells with environmental risk entropy higher than the preset environmental risk entropy, and load a preset texture-simplified physical model onto grid cells with environmental risk entropy lower than the preset environmental risk entropy to obtain a digital twin model.

[0069] It should be noted that the preset environmental risk entropy is a pre-set critical value used to distinguish between high-risk and low-risk levels.

[0070] By querying the environmental risk topology map generated in step S13, the environmental risk entropy corresponding to the center point or coverage area of ​​each grid cell in the multi-resolution three-dimensional grid can be directly obtained, thereby determining the environmental risk entropy of each grid cell.

[0071] During model loading, the environmental risk entropy of each mesh cell is compared with the preset environmental risk entropy. For mesh cells with an environmental risk entropy higher than the preset environmental risk entropy, a high-precision texture map is called from the model library and bound to a complex physical behavior model, i.e., a preset texture-physical behavior model (such as detailed interactions such as simulating equipment collisions and fluid diffusion), to realistically reproduce the details of the high-risk environment. For mesh cells with an environmental risk entropy lower than the preset environmental risk entropy, a simplified texture is loaded and associated with a basic physical model, i.e., a preset texture-simplified physical model (such as containing only simple collision detection), to quickly render low-risk areas. This achieves adaptive construction of the digital twin model, ensuring high simulation fidelity and interactive realism in high-risk areas while effectively optimizing the overall model's resource consumption and significantly improving the running efficiency and real-time interactive experience of subsequent workplace injury prevention training simulations.

[0072] Furthermore, step S12 may include steps S121 to S123:

[0073] Step S121: Normalize the corresponding environmental data streams through each edge computing node to generate multiple environmental parameter vectors.

[0074] Using preset parameter ranges (such as historical minimum and maximum values ​​of temperature, humidity, light intensity, and gas concentration), the minimum-maximum normalization algorithm is used to linearly transform the value of each environmental parameter in the environmental data stream to the [0,1] interval. For example, for the temperature parameter, the normalized value is calculated as (current temperature value - historical minimum temperature) / (historical maximum temperature - historical minimum temperature), thereby eliminating the dimensional differences between different environmental parameters. That is, the normalization operation in this embodiment is to divide the difference between the current value and the historical minimum value by the difference between the historical maximum value and the historical minimum value.

[0075] After processing, each edge computing node combines the values ​​of the normalized environmental data stream at the same time into a standardized vector, namely the environmental parameter vector, so that environmental data of different units and magnitudes are comparable and consistent, providing a standardized data foundation for subsequent calculation of environmental fluctuation characteristics and environmental risk entropy.

[0076] Step S122: Calculate the fluctuation amplitude of each environmental parameter vector in each environmental dimension within the sliding time window to obtain multiple environmental fluctuation characteristics.

[0077] By setting a sliding time window of fixed length, the time series data of each environmental parameter vector within the sliding time window are extracted sequentially. The variance of the time series data of each environmental dimension (such as temperature, humidity, light intensity, and gas concentration) within the sliding time window is calculated as a quantitative indicator of its corresponding fluctuation amplitude. This yields the environmental fluctuation characteristics corresponding to each environmental dimension, transforming the normalized environmental parameter vector into a feature sequence that dynamically reflects the intensity of fluctuations of each environmental parameter in the near future. This captures the instability and changing trends of the environmental state, providing a crucial dynamic input basis for calculating the uncertainty measure that characterizes the comprehensive risk in subsequent steps.

[0078] Step S123: Based on the characteristics of each environmental fluctuation, calculate the uncertainty measure of each edge computing node in each environmental dimension, and then fuse the uncertainty measures in each environmental dimension to generate the environmental risk entropy corresponding to each edge computing node.

[0079] The environmental fluctuation characteristics calculated within the sliding time window for each environmental dimension are used as the basic data for the state uncertainty of the corresponding environmental dimension, and the information entropy calculation formula is applied to quantify them. For example, for the temperature dimension, its uncertainty measure is calculated using the following formula 1:

[0080] ——Formula 1

[0081] in, This represents the probability that the normalized temperature fluctuation amplitude occurs within the discretized interval. The larger the temperature fluctuation amplitude, the more unpredictable the temperature fluctuation. This represents the uncertainty measure in the temperature dimension. Uncertainty measures in other dimensions are also calculated using the formula shown in Formula 1. By substituting the normalized humidity fluctuation range, temperature fluctuation range, and gas concentration range into Formula 1, the corresponding uncertainty measure can be calculated.

[0082] After obtaining the uncertainty measures for the four environmental dimensions—temperature, humidity, light intensity, and gas concentration—a weighted summation method is used to fuse these uncertainty measures. Specifically, different weighting coefficients are preset based on the degree of influence of each environmental dimension on work-related injury risk. Finally, the weighted sum is calculated as the environmental risk entropy of the corresponding edge computing node. The calculation formula is: Environmental Risk Entropy. Among them, w1, w2, w3 and w4 are the weighting coefficients of the four ambient temperatures; A measure of uncertainty in the humidity dimension; A measure of uncertainty in the illumination dimension; As a measure of uncertainty in gas concentration, it integrates the uncertainty measures of multiple environmental dimensions into a single and quantified environmental risk entropy. This environmental risk entropy not only reflects the overall instability of the local environment, but also reflects the differences in the contribution of different environmental factors to the risk.

[0083] In one feasible implementation, refer to Figure 3 As shown, the specific execution steps of the simulation module include steps S21 to S24:

[0084] Step S21: Present a virtual scene of the venue to the user through a digital twin model.

[0085] The digital twin model presents a virtual scene of the workplace to users. Specifically, it is achieved by using terminal devices, such as computer monitors, to load multi-resolution 3D meshes generated by building modules and recorded texture and physical behavior models. Graphics rendering technology is then used to calculate and draw a 3D visualization scene consistent with the real work site in real time. This transforms the digital twin model into a virtual scene that users can perceive, providing them with an immersive and highly realistic interactive training platform. This allows users to intuitively familiarize themselves with the layout, equipment, and potential hazards of the work site without physical risks, thus laying a safe and effective visual and interactive foundation for subsequent simulation operations.

[0086] Step S22: Based on the user's real-time operation in the virtual scene of the site, simulation is performed using a digital twin model to obtain the environmental state changes and equipment interaction feedback caused by the real-time operation.

[0087] By continuously monitoring user input commands on terminal devices and mapping these commands to corresponding actions of objects in a virtual scene, the digital twin model then calculates the actions based on its built-in physical rules and logical states. For example, when a user operates a virtual valve, the model calculates the flow and pressure changes of the medium in the pipeline in real time based on fluid dynamics equations and updates the display status of related equipment. Or, when a user approaches a high-risk area, the model triggers corresponding audible and visual alarm feedback based on the behavior model loaded in that grid cell. This creates a dynamic simulation environment that can respond to user behavior in real time and accurately, enabling user operations to immediately trigger visualized changes in environmental state and device interaction feedback. This provides users with an immersive interactive experience that allows them to verify the consequences of their operations and perceive the causal relationship between their operations and risks.

[0088] Step S23: Collect user decision-making actions under changes in environmental conditions and device interaction feedback.

[0089] After the digital twin model triggers environmental state updates and device feedback, it records the user's subsequent operation command sequence on the terminal device, such as keyboard operations and mouse click trajectories, and simultaneously collects physiological response data obtained through wearable devices, such as eye movement trajectories and heart rate variability. This allows for the complete capture of the user's cognitive judgments and operational responses when facing a dynamic simulation scenario. By acquiring multimodal data sequences that reflect the user's real decision-making logic and stress response, it provides the original basis for characterizing individual behavioral features for subsequent quantitative operational analysis, thereby establishing a complete data closed loop from virtual scene feedback to user behavioral response.

[0090] Step S24: Quantify the decision-making actions to obtain behavioral data.

[0091] The decision action sequence collected in step S23 is divided into preset time windows. The action type within each time window is one-hot encoded, the action duration is quantized in milliseconds, and the operation trajectory coordinates are serialized and recorded. These values ​​are then aligned and integrated into a structured data table by timestamps. This transforms continuous decision actions into discrete, numerical behavioral data, realizing the transformation of unstructured user operations into machine-readable data format.

[0092] In one feasible implementation, refer to Figure 4 As shown, the specific execution steps of the analysis module include steps S31 to S34:

[0093] Step S31: Extract features from the behavioral data to obtain behavioral feature vectors.

[0094] By employing a multimodal feature fusion method, representative features are extracted from the operation command sequences and physiological response data of behavioral data and then fused. Specifically, for keyboard operation sequences, a sliding window is used to statistically analyze the keystroke frequency and combination patterns per unit time; for mouse click trajectories, the curvature change of the movement trajectory and the click distribution heatmap are calculated; for eye movement trajectories, the gaze point distribution entropy and saccade velocity variance are extracted; and for heart rate variability, the time domain standard deviation and frequency domain high-low power ratio are calculated. Finally, the feature vectors of the operation command sequences (such as keystroke frequency and mouse trajectory curvature) and the feature vectors of the physiological response data (such as eye movement entropy and heart rate variability index) are directly concatenated to form a unified behavioral feature vector, thereby realizing the fusion of discrete behavioral data into a comprehensive feature representation that can fully reflect the user's operating habits and physiological state.

[0095] Step S32: Calculate the behavioral feature vector using a preset safety assessment model to obtain the safety deviation.

[0096] It should be noted that the preset security assessment model is a machine learning model pre-trained based on a large amount of standard security behavior data. By learning the distribution pattern of security behavior feature vectors, the preset security assessment model establishes the decision boundary between normal operation and risky operation.

[0097] During calculation, the behavioral feature vector is input into the preset security assessment model. The preset security assessment model processes the behavioral feature vector through forward propagation or kernel function transformation and outputs a continuous value. This continuous value represents the degree of deviation between the current user's behavioral feature vector and the standard safe behavior pattern in the feature space, i.e., the security deviation degree. In this way, the user's complex behavioral characteristics are transformed into a unified and quantifiable security indicator, which objectively reflects the degree of difference between the user's operation and the security specifications.

[0098] Step S33: Input the safety deviation into the preset risk assessment rules for risk matching to obtain the risk type and work injury risk level.

[0099] It should be noted that the preset risk assessment rules are an if-then rule base built on domain expert knowledge. It clearly defines the standards for classifying work injury risk levels corresponding to different safety deviation ranges. For example, it classifies a safety deviation of 0-0.2 as low work injury risk, 0.2-0.5 as medium work injury risk, and above 0.5 as high work injury risk. It also classifies the risk types corresponding to various operational behavior patterns, such as typical risk patterns such as incorrect operation sequence and excessive response delay.

[0100] When performing risk matching, the safety deviation is compared with the preset threshold range in the preset risk assessment rule base to determine its corresponding work injury risk level. At the same time, abnormal combination patterns of specific features in the behavioral feature vector, such as high-frequency operation combined with abnormal eye movement trajectory, are matched with the risk feature patterns defined in the preset risk assessment rule base to determine the specific risk type. This transforms the quantified safety deviation into a risk assessment conclusion with clear semantic interpretation, enabling the system to not only judge the magnitude of the risk but also identify its nature, providing a basis for classification and grading decisions to generate targeted work injury early warning information.

[0101] Step S34: Generate work injury early warning information based on risk type and work injury risk level.

[0102] By querying the pre-defined warning information template library, the library predefines corresponding risk description text, severity indicators, and improvement measure suggestions for each risk type and work injury risk level combination.

[0103] Based on the risk type and work injury risk level, the system matches the corresponding template from the preset early warning information template library and automatically fills in specific parameters, such as risk location and deviation value, to generate a structured work injury early warning information. This work injury early warning information typically includes risk classification, risk level, specific manifestations, and recommended actions, thereby transforming abstract risk assessment results into personalized early warning feedback that users can intuitively understand and that has operational guidance significance.

[0104] In one feasible implementation, refer to Figure 5 As shown, the specific execution steps of the output module include steps S41 to S44:

[0105] Step S41: Perform data quantification processing on work injury early warning information and historical training records to obtain safety assessment data.

[0106] The purpose of quantifying work injury early warning information and historical training records is to transform discrete safety information of different types and sources into a unified numerical expression, thereby providing a data foundation for subsequent safety credit score calculation.

[0107] Step S42: Based on the safety assessment data, assign weights to the work injury early warning information and historical training records to obtain a weighted assessment value.

[0108] A weighting rule is established based on the time attributes and risk severity of safety assessment data. First, a time decay function is applied to historical training records to give higher weight to recent training performance. Simultaneously, differentiated weighting coefficients are set according to the risk level of work injury warning information, with higher-risk warnings corresponding to higher weights. Then, each type of safety assessment data is multiplied by its corresponding weight, and finally, a weighted sum is calculated to obtain a comprehensive weighted assessment value reflecting the user's recent safety performance. This ensures that recent performance and serious risky behaviors have a greater impact on the score result, thereby ensuring that the safety credit score can promptly reflect changes in the user's current safety status.

[0109] Step S43: Calculate the user's score within their respective user group based on the weighted evaluation value to obtain a relative safety score.

[0110] First, the weighted assessment values ​​of all users in the same group, such as those in the same position or department, are collected. Then, a percentile ranking algorithm is used to compare the current user's weighted assessment value with the weighted assessment values ​​of other users in the group, and calculate the user's specific percentile ranking in the group. For example, if a user's weighted assessment value exceeds 80% of the users in the group, their relative safety score is recorded as 80. In this way, the user's absolute safety performance is transformed into a relative ranking in a specific reference group, effectively eliminating the impact of fluctuations in the overall level of the group or differences in risk between different positions on the assessment results, thereby ensuring that the final safety credit score can reflect individual performance.

[0111] Step S44: Map the relative security score to a preset score range to obtain a security credit score.

[0112] The relative safety score, based on a percentage system, is mapped proportionally to a preset score range using a linear transformation function. For example, when the preset score range is 0-100 points, the formula "Safety Credit Score = Relative Safety Score" is used for direct mapping. Alternatively, when the preset score range is 0-1000 points, the formula "Safety Credit Score = Relative Safety Score × 10" is used for linear amplification, ensuring a strict linear correspondence between the score and the score.

[0113] Furthermore, step S41 may include steps S411 to S413:

[0114] Step S411: By using a pre-set quantitative indicator system, the number of warnings, warning levels, training results, and training completion are processed with unified dimensions to obtain a set of quantitative indicators.

[0115] The number of warnings is converted into a value in the 0-1 range using a linear normalization method. Different warning levels (such as high, medium, and low) are mapped to corresponding numerical coefficients (such as 1.0, 0.6, and 0.2). The percentage-based training scores are converted into standard scores, and the training completion rate is directly quantified as a percentage of completion.

[0116] Through the above-mentioned rule-based processing, all indicators are uniformly transformed into comparable dimensionless values, forming a set of quantitative indicators. This eliminates the differences in dimensions and magnitudes between indicators from different sources, enabling multi-source heterogeneous data such as work injury early warning information and historical training records to be comprehensively calculated under the same standard, providing a fair and reliable numerical basis for subsequent weighted fusion.

[0117] Step S412: Weighted fusion of the quantitative indicators in the quantitative indicator set to generate a comprehensive security assessment value.

[0118] By employing a pre-defined weighting scheme, each quantitative indicator in the set of quantitative indicators—namely, the number of warnings, the warning level, training performance, and training completion rate—is assigned a corresponding weight coefficient. These weight coefficients are set based on the importance of each quantitative indicator to the overall security assessment. Then, a weighted average algorithm is used to multiply the value of each quantitative indicator by its corresponding weight coefficient, sum the results, and divide by the sum of the weight coefficients to obtain the comprehensive security assessment value. This process combines multiple independent quantitative indicators according to their importance into a unified comprehensive score, thereby comprehensively and reasonably reflecting the user's overall security status.

[0119] Step S413: Map the comprehensive security assessment value to a preset security assessment range to obtain security assessment data.

[0120] By using a linear transformation function, the comprehensive security assessment value is scaled proportionally to the minimum and maximum values ​​of the preset security assessment range. For example, if the original range of the comprehensive security assessment value is [0, 100], while the preset security assessment range is [0, 10], then the formula "security assessment data = (comprehensive security assessment value / 100) × 10" is used to calculate the comprehensive security assessment value, thereby mapping the comprehensive security assessment value to the preset security assessment range. This standardizes the comprehensive security assessment value to a unified numerical range, eliminates scale differences between different assessment values, and ensures that all data have consistent comparability and operability.

[0121] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A workplace injury prevention training system, characterized in that, include: A construction module is used to acquire environmental data streams from the actual work site and construct a digital twin model corresponding to the actual work site based on the environmental data streams. The simulation module is used to conduct work injury prevention training for users through the digital twin model, and to collect user behavior data during the work injury prevention training process; The analysis module is used to perform work injury analysis based on the behavioral data and generate work injury early warning information; The output module is used to generate a safety credit score based on the work injury early warning information and historical training records; The step of acquiring environmental data streams from the actual work site and constructing a digital twin model corresponding to the actual work site based on the environmental data streams includes: By acquiring the environmental data stream at each of the sensor network points deployed on the actual work site; Based on the edge computing nodes embedded in each of the sensor network points, the environmental data streams are processed to generate multiple environmental risk entropies. Based on the multiple environmental risk entropies, an environmental risk topology map of the actual work site is generated; Based on the environmental risk entropy in the environmental risk topology map, a grid resolution division threshold is set to re-divide the preset three-dimensional network of the real work site to obtain a multi-resolution three-dimensional grid. The environmental risk entropy of each grid cell in the multi-resolution 3D mesh is determined. A preset texture-physical behavior model is loaded on the grid cells whose environmental risk entropy is higher than the preset environmental risk entropy, and a preset texture-simplified physical model is loaded on the grid cells whose environmental risk entropy is lower than the preset environmental risk entropy, to obtain the digital twin model. The step of processing each environmental data stream based on edge computing nodes embedded in each of the sensor network points to generate multiple environmental risk entropies includes: Each edge computing node performs normalization processing on the corresponding environmental data stream to generate multiple environmental parameter vectors. Calculate the fluctuation amplitude of each environmental parameter vector in each environmental dimension within the sliding time window to obtain multiple environmental fluctuation characteristics; Based on the environmental fluctuation characteristics, the uncertainty measure of each edge computing node in each environmental dimension is calculated, and the uncertainty measures in each environmental dimension are fused to generate the environmental risk entropy corresponding to each edge computing node.

2. The work injury prevention training system according to claim 1, characterized in that, The steps of conducting workplace injury prevention training for users through the digital twin model and collecting user behavioral data during the workplace injury prevention training process include: The digital twin model is used to present a virtual scene of the venue to the user; Based on the user's real-time operations in the virtual scene of the venue, the digital twin model is used for simulation to obtain the environmental state changes and device interaction feedback caused by the real-time operations. Collect user decision-making actions in response to changes in the environmental state and feedback from device interactions; The decision-making action is quantified to obtain the behavioral data.

3. The work injury prevention training system according to claim 1, characterized in that, The steps of performing work injury analysis based on the behavioral data and generating work injury early warning information include: Feature extraction is performed on the behavioral data to obtain a behavioral feature vector; The safety deviation is obtained by calculating the behavioral feature vector using a preset safety assessment model. The safety deviation is input into a preset risk assessment rule for risk matching to obtain the risk type and work injury risk level; The work injury early warning information is generated based on the risk type and the work injury risk level.

4. The work injury prevention training system according to claim 1, characterized in that, The step of generating a safety credit score based on the work injury early warning information and historical training records includes: The work injury early warning information and the historical training records are subjected to data quantification processing to obtain safety assessment data; Based on the safety assessment data, the work injury early warning information and the historical training records are weighted and assigned to obtain a weighted assessment value. Based on the weighted evaluation value, the user's score within their respective user group is calculated to obtain a relative security score; The relative security score is mapped to a preset score range to obtain the security credit score.

5. The work injury prevention training system according to claim 4, characterized in that, The work injury early warning information includes the number of early warnings and the early warning level; the historical training records include training performance and training completion rate; the step of quantifying the work injury early warning information and the historical training records to obtain safety assessment data includes: By using a pre-defined quantitative indicator system, the number of warnings, the warning level, the training results, and the training completion rate are processed with unified dimensions to obtain a set of quantitative indicators. The quantitative indicators in the set of quantitative indicators are weighted and fused to generate a comprehensive security assessment value; The comprehensive security assessment value is mapped to a preset security assessment range to obtain the security assessment data.

Images