A mine safety training method and system based on digital twinning and virtual reality

Abstract

The application provides a mine safety training method and system based on digital twinning and virtual reality, the method comprising: acquiring laser scanning point cloud and geological exploration data of a mine underground tunnel, and constructing a high-precision virtual underground environment model based on the data; based on the virtual underground environment model, combining explosion condition judgment, triggering disaster simulation, and using real-time rendering and dynamic labeling technology to generate a virtual training environment; acquiring operation behavior data of a training personnel in the virtual training environment, and evaluating the operation behavior data to obtain a behavior evaluation result; based on the behavior evaluation result and environment parameter change data, using machine learning to construct and train an evaluation model, combining different mine area geological conditions and mining process characteristics to generate customized training content. The method can construct a high-reduction virtual scene, solve the problem of single traditional training scene, realize quantitative evaluation of training effect, and improve the pertinence and effectiveness of training.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mine safety training, and particularly relates to a mine safety training method and system based on digital twinning and virtual reality. BACKGROUND

[0002] With the expansion of mining scale and the increase of operation complexity, the traditional mine safety training mode has problems such as single scene, insufficient risk simulation, and difficult quantification of training effect, and cannot meet the training needs of modern mines for safety skills and emergency capabilities of high-risk operation personnel. Digital twinning technology can accurately reproduce the dynamic changes of complex scenes by constructing a virtual mapping model of physical entities; virtual reality technology provides hardware and software support for immersive interactive training. However, the existing technology still has technical bottlenecks in multi-source heterogeneous data fusion, high-precision environment modeling, dynamic disaster simulation, and personalized training system construction, and it is difficult to realize the closed-loop optimization of mine operation whole-process risk simulation and training effect. SUMMARY

[0003] (I) Technical problems to be solved

[0004] Therefore, the present application provides a mine safety training method and system based on digital twinning and virtual reality to solve the problems of single training scene, insufficient risk simulation, and difficult quantification of training effect in the background art.

[0005] (II) Technical solutions

[0006] In order to achieve the above purpose, the present application provides a mine safety training method based on digital twinning and virtual reality, comprising:

[0007] S1: Obtain laser scanning point cloud and geological exploration data of a mine underground roadway, and construct a high-precision virtual underground environment model based on the data using a three-dimensional reconstruction algorithm;

[0008] S2: Based on the virtual underground environment model, combined with explosion condition judgment, trigger disaster simulation, and generate a virtual training environment using real-time rendering and dynamic labeling technology; specifically comprising:

[0009] S201: Based on the virtual underground environment model, construct a multi-physical field coupling simulation model including airflow, gas concentration, toxic gas concentration, temperature, etc. in the roadway;

[0010] S202: Solve the airflow field distribution in the roadway using a computational fluid dynamics algorithm, and calculate the time-space evolution of gas concentration combined with a reaction kinetics model;

[0011] S203: Apply dynamic boundary conditions to the multi-physical field coupling simulation model, combine the spatio-temporal evolution of gas concentration, and use machine learning algorithms to optimize the explosion conditions;

[0012] S204: Real-time monitoring of environmental parameters in the multi-physical field coupling simulation model, triggering disaster simulation when explosion conditions are met, starting shock wave propagation numerical calculation, and obtaining propagation simulation results;

[0013] S205: Based on the propagation simulation results, use a physically driven particle system to generate dynamic effects such as explosion flames and smoke diffusion, combine the shock wave attenuation law calculated by the sound wave equation, and construct multi-channel stereo sound field data;

[0014] S206: Calculate the regional risk probability based on the gas concentration, toxic gas diffusion range, temperature field distribution, and roadway structure integrity in the multi-physical field coupling simulation model, divide the risk level, and generate a time-space dynamic change risk thermodynamic map;

[0015] The calculation formula of the regional risk probability is as follows:

[0016]

[0017] Where (x, y, z) represents the spatial coordinates, t represents the time; P risk (x, y, z, t) represents the regional risk probability; f i (x, y, z, t) represents the i, i∈{1, 2, 3, 4} physical field risk function, including gas concentration, toxic gas concentration, temperature and structure damage; ω i (t) represents the dynamic weight coefficient of the i-th physical field; g(t) represents the space-time decay function; f gas (x, y, z, t) represents the gas concentration function; C(x, y, z, t) represents the gas concentration at coordinates (x, y, z) at time t; C exp represents the explosive limit concentration; C low represents the lower threshold; α represents the sensitivity coefficient; M represents the indicator function; f tox (x, y, z, t) represents the toxic gas function; D(x, y, z, t) represents the comprehensive toxicity index at coordinates (x, y, z) at time t; D th represents the risk threshold; β represents the control probability mutation slope; f temp (x, y, z, t) represents the temperature field function; T(x, y, z, t) represents the temperature at coordinates (x, y, z) at time t; T0 represents the environmental temperature reference value; T eirt represents the critical temperature; f struct(x, y, z, t) represents a structure damage function; σ(x, y, z, t) represents a structure stress at the time t at the coordinates (x, y, z); σ y represents a yield strength; γ represents a material sensitivity coefficient;

[0018] S207: generating a virtual training environment by using real-time rendering and dynamic labeling technology according to the stereo sound field data and the dangerous thermal diagram;

[0019] S3: obtaining operation behavior data of the training personnel in the virtual training environment, and evaluating the operation behavior data to obtain a behavior evaluation result;

[0020] S4: constructing and training an evaluation model by using machine learning based on the behavior evaluation result and the environment parameter change data, and generating customized training content in combination with different mine geological conditions and mining process characteristics.

[0021] Preferably, the space-time attenuation function g(t) is represented as follows:

[0022]

[0023] wherein t0 represents a time when a disaster occurs; t max represents a duration; d(x, y, z) represents a distance to a disaster source; d0 represents a characteristic distance, which is a key parameter for measuring a spatial attenuation speed of disaster influence, and when the distance d(x, y, z) to the disaster source is equal to d0, the spatial attenuation factor of the disaster influence is e -1 .

[0024] Preferably, the dynamic weight coefficient ω i (t) of the i-th physical field is represented as follows:

[0025]

[0026] wherein λ i represents an influence factor of the i-th physical field.

[0027] Preferably, S1 specifically comprises:

[0028] S101: obtaining laser scanning point clouds and geological exploration data of a mine underground roadway, removing redundancies and noise points in the data, and obtaining standardized pretreatment data corresponding to a real mine by using a data registration technology to unify a coordinate system;

[0029] S102: discretizing continuous space by using a voxelization algorithm, and obtaining a structured data grid by combining a space subdivision technique to divide regions;

[0030] S103: Based on the structured data grid, a point cloud surface reconstruction algorithm is used to generate a basic three-dimensional model framework, and through splicing technology, the preliminary environmental three-dimensional model is obtained by integrating each region;

[0031] S104: The preliminary environmental three-dimensional model is given a model surface material texture using a texture mapping algorithm, and combined with detail enhancement technology and micro-terrain optimization technology, an underground environment model is generated;

[0032] S105: Based on the underground environment model, the spatial position of the equipment is determined using spatial annotation technology, the training information is semantically associated with the spatial position of the equipment, and a virtual underground environment model is obtained.

[0033] Preferably, S3 specifically comprises:

[0034] S301: Collect the body movements, eye tracking and equipment operation data of the training personnel in the virtual training environment to obtain the original behavior data set;

[0035] S302: Data cleaning is performed on the original behavior data set, abnormal values are removed by a noise reduction algorithm, and data format is unified by data standardization technology to obtain preprocessed behavior data;

[0036] S303: Based on the preprocessed behavior data, a spatio-temporal graph convolution network is used to extract the spatio-temporal features of the action sequence, and a attention mechanism is used to strengthen the key operation behavior features to obtain a behavior feature vector;

[0037] S304: The behavior feature vector is evaluated using an evaluation algorithm based on the fusion of rule engine and machine learning, and the behavior evaluation indicators such as operation compliance and response time are calculated by comparing with the pre-set safety operation specification and emergency handling process;

[0038] S305: According to the behavior evaluation indicators, a clustering algorithm is used to classify the behavior patterns of the training personnel, the weights of each indicator are determined by the analytic hierarchy process, and the behavior evaluation results are constructed.

[0039] Preferably, S4 specifically comprises:

[0040] S401: Based on the behavior evaluation indicators and environmental parameter change data, feature engineering technology is used to filter and reduce the data, and a high correlation training data set is obtained by combining principal component analysis algorithm;

[0041] S402: A combination model of random forest and gradient boosting tree is used as an initial evaluation model, and the training data set is used for training;

[0042] S403: According to different geological conditions and mining process characteristics of different mining areas, knowledge graph technology is used to extract key features, and adaptive weight adjustment algorithm is used to dynamically optimize the scoring weights of the initial evaluation model;

[0043] S404: The optimized evaluation model is simulated by a reinforcement learning algorithm to obtain feedback in different training scenarios, and parameters are adjusted to obtain a refined evaluation model;

[0044] S405: Based on the output result of the refined evaluation model, natural language generation technology and interactive scene construction algorithm are used to generate customized training content including theoretical explanation, practical simulation and emergency drill.

[0045] On the other hand, the present application also provides a mine safety training system based on digital twin and virtual reality, which comprises a data acquisition modeling module, a disaster simulation generation module, a behavior acquisition evaluation module and a training content generation module.

[0046] The data acquisition modeling module is used to obtain laser scanning point cloud data and geological exploration data of underground mine roadway, and a high-precision virtual underground environment model is constructed based on the data using a three-dimensional reconstruction algorithm, and the virtual underground environment model is transmitted to the disaster simulation generation module.

[0047] The disaster simulation generation module is used to trigger disaster simulation based on the virtual underground environment model combined with explosion condition judgment algorithm, generate virtual training environment using real-time rendering and dynamic labeling technology, and transmit the virtual training environment to the behavior acquisition evaluation module.

[0048] The behavior acquisition evaluation module is used to obtain operation behavior data of training personnel in the virtual training environment, evaluate the operation behavior data using behavior analysis algorithm to obtain behavior evaluation results, and transmit the behavior evaluation results to the training content generation module.

[0049] The training content generation module is used to train individualized evaluation model based on the behavior evaluation results and environment parameter change data using machine learning algorithm, generate customized training content combined with different mine geological conditions and mining process characteristics.

[0050] (Three) beneficial effects

[0051] From the above technical solution, the mine safety training method and system based on digital twin and virtual reality proposed by the present application have the following beneficial effects:

[0052] 1. The virtual underground environment model of the present application is highly consistent with the real mine environment in terms of spatial structure, material performance and semantic information, providing reliable data basis for disaster simulation and safety training. When training personnel interact with equipment in the virtual environment, they can obtain corresponding training information in real time, enhance the pertinence and effectiveness of training, realize the deep integration of scene and knowledge, improve the effect of safety training, and meet the needs of mine safety training for scene reality and information integrity.

[0053] 2. Through multi-physical field coupling modeling and optimization, the occurrence and propagation process of downhole disasters can be simulated realistically; combined with dynamic rendering and risk assessment, the simulation results are converted into a visual and interactive virtual training environment. Training personnel can master the risk distribution law under different disaster scenarios and strengthen their risk identification ability; the dynamically updated risk level can simulate the disaster evolution process to assist training personnel in learning emergency escape path planning and decision logic, so that safety training is transformed from static knowledge infusion to dynamic risk response operation, significantly improving the authenticity and practicality of training content.

[0054] 3. The behavior details of the training personnel can be accurately captured, and the operation standardization and emergency disposal ability can be objectively quantified. Compared with the traditional manual evaluation method, the present application can effectively reduce subjective errors and improve evaluation efficiency and accuracy. Based on clustering and weight analysis, the evaluation results can clearly show the individual behavior differences and provide data support for personalized training program development.

[0055] 4. The correlation characteristics of the training personnel behavior and environmental parameters can be deeply mined to accurately evaluate individual training needs; the evaluation model is optimized based on the actual characteristics of the mine area to ensure that the training content meets the real needs of different operation scenarios. Compared with the traditional fixed training mode, the adaptability and effectiveness of the training content can be significantly improved to help training personnel efficiently master the corresponding skills while reducing the enterprise's repeated training costs. BRIEF DESCRIPTION OF DRAWINGS

[0056] The features and advantages of the present application will be more clearly understood through reference to the accompanying drawings, which are schematic and are not intended to be limiting of the present application, in which:

[0057] Figure 1 The structure diagram of the mine safety training system based on digital twinning and virtual reality of the embodiment of the present application. DETAILED DESCRIPTION

[0058] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0059] Embodiment one:

[0060] The present application provides a mine safety training method based on digital twinning and virtual reality, comprising:

[0061] S1: Obtain laser scanning point cloud and geological exploration data of underground mine roadway, and construct a high-precision virtual underground environment model based on the data using a three-dimensional reconstruction algorithm; specifically comprising:

[0062] S101: Obtain laser scanning point cloud and geological exploration data of underground mine roadway, remove redundancy and noise in the data, unify the coordinate system through data registration technology, and obtain standardized pretreated data corresponding to the real mine;

[0063] Specifically, point cloud information and geological exploration data of underground mine roadway are collected by terminals such as laser scanners and geological exploration equipment, and the data includes information such as roadway space form, rock structure, and equipment distribution. Among them, the equipment includes various actual equipment in the underground environment, such as mining equipment, ventilation equipment, and transportation equipment. After collection, the data is pretreated by denoising, registration, and other pretreatment, and the data coordinate system is unified to eliminate redundant information.

[0064] S102: Discretize the continuous space using a voxelization algorithm on the standardized pretreated data, and divide the data area using a space subdivision technique to obtain a structured data grid;

[0065] S103: Based on the structured data grid, generate a basic three-dimensional model framework using a point cloud surface reconstruction algorithm, integrate each data area through splicing technology, and obtain a preliminary environment three-dimensional model;

[0066] Specifically, after obtaining the structured data grid, the discrete point cloud data needs to be converted into a continuous three-dimensional surface model. The point cloud surface reconstruction algorithm analyzes the spatial position relationship of each point in the grid, uses the moving least squares method (MLS) or implicit surface fitting method to establish geometric connection between point cloud data, and generates a smooth surface model. This process needs to optimize parameters such as data density and surface curvature to ensure that the reconstructed surface fits the real roadway form and geological structure. Since the mine scene is large in scale, the data is usually collected and processed by area, so model splicing technology is needed to integrate each sub-area model. Through feature matching algorithm, the corresponding points or surfaces of adjacent model boundaries are identified, the sub-models are aligned based on coordinate transformation, and the splicing gaps are eliminated through triangular mesh fusion technology, and finally the preliminary model covering the entire underground environment (i.e. preliminary environment three-dimensional model) is constructed.

[0067] S104: Assign surface material texture to the preliminary environment three-dimensional model using a texture mapping algorithm, combine detail enhancement technology and micro-terrain optimization technology, and generate an underground environment model;

[0068] S105: Based on the underground environment model, determine the spatial position of the equipment using spatial labeling technology, semantically associate the training information with the spatial position of the equipment, and obtain a virtual underground environment model;

[0069] Specifically, the spatial annotation technology is used to determine the device coordinates, and the training information such as the operation specification, safety precautions, and emergency handling process of the device is semantically associated with the spatial position of the device, so that the virtual underground environment model not only has geometric shape (i.e., spatial topology) and physical properties (including devices and environment), but also carries the knowledge content required for training, and completes the construction of the virtual underground environment model.

[0070] The embodiment realizes the systematic conversion from the real data of the mine underground tunnel to the virtual underground environment model. Through the fusion of multiple technologies, it ensures that the virtual underground environment model is highly consistent with the real mine environment in terms of spatial structure, material performance, and semantic information, and provides a reliable data basis for disaster simulation and safety training. Data standardization and structured processing improve the model construction efficiency, and the semantic association of training information enhances the functionality of the model. When the training personnel interact with the device in the virtual environment, they can obtain the corresponding training information in real time, enhance the pertinence and effectiveness of training, realize the deep integration of scene and knowledge, improve the effect of safety training, and meet the needs of mine safety training for scene reality and information integrity.

[0071] S2: Based on the virtual underground environment model, combined with the explosion condition judgment, trigger disaster simulation, and generate a virtual training environment by using real-time rendering and dynamic annotation technology; specifically including:

[0072] S201: Based on the virtual underground environment model, construct a multi-physical field coupling simulation model including airflow, gas concentration, toxic gas concentration, temperature, and other physical fields in the tunnel;

[0073] Specifically, the multi-physical field coupling simulation model is constructed based on the spatial layout of the virtual underground environment model, ventilation parameters, and device physical properties, and is a model used to simulate complex physical processes in the underground. The model couples multiple physical fields such as fluid mechanics and reaction kinetics, uses computational fluid dynamics algorithms to solve the airflow field distribution in the tunnel, combines with the reaction kinetics model to calculate the spatiotemporal evolution of gas concentration, updates the environmental parameters in real time by applying dynamic boundary conditions (such as changes in ventilation parameters, fluctuations in gas emission, etc.), and uses machine learning algorithms to optimize the explosion conditions, to realize the accurate simulation of the occurrence and propagation process of disasters such as gas explosion.

[0074] S202: Use computational fluid dynamics algorithms to solve the airflow field distribution in the tunnel, and combine with the reaction kinetics model to calculate the spatiotemporal evolution of gas concentration;

[0075] Preferably, the computational fluid dynamics algorithm solves the Navier-Stokes equations by discretizing the roadway space, combining the roadway boundary conditions and ventilation parameters to obtain the velocity and pressure distribution of the airflow in the roadway. This process can consider the air volume and air pressure of the ventilation equipment, as well as the frictional resistance of the roadway wall surface, to ensure the accuracy of the airflow field simulation. At the same time, the diffusion and convection processes of gas are coupled with chemical reactions based on the reaction kinetics model. Based on the law of conservation of mass, partial differential equations are established for the time and space variation of gas concentration, considering factors such as gas emission, dilution, and chemical reaction consumption. The equations are solved by numerical methods to obtain the time and space evolution data of gas concentration in the roadway. This data reflects the changes in gas concentration at different times and different locations.

[0076] S203: Apply dynamic boundary conditions to the multi-physical field coupling simulation model, combine the time and space evolution of gas concentration, and use machine learning algorithms to optimize the explosion conditions;

[0077] Preferably, the dynamic boundary conditions include real-time changes in ventilation parameters (such as air volume and air pressure), the running state of equipment in the roadway (such as fan start-stop and air door opening-closing), dynamic fluctuations in gas emission, and external environmental disturbances (such as temperature and pressure changes). These conditions will be dynamically adjusted over time, affecting the calculation process of the multi-physical field coupling simulation model. Optimizing the explosion conditions involves using machine learning algorithms to intelligently calibrate and adjust the critical conditions for explosion (such as gas concentration threshold, oxygen content, and ignition energy) based on the time and space evolution data of gas concentration and the simulation results of the multi-physical field coupling simulation model, making them more suitable for the complex scenarios of actual underground disasters, thereby improving the accuracy and reliability of the explosion simulation trigger.

[0078] S204: Real-time monitoring of environmental parameters in the multi-physical field coupling simulation model, triggering disaster simulation when the explosion conditions are met, starting the numerical calculation of shock wave propagation to obtain the propagation simulation results;

[0079] Further, the system constructs a multi-physical field coupling simulation model based on the disaster simulation calculation unit, and monitors the gas concentration, oxygen content, temperature, and pressure in the virtual underground environment in real time. When these parameters meet the explosion conditions optimized by the machine learning algorithm (such as reaching the explosion limit of gas concentration and the presence of an ignition source), the system triggers disaster simulation, starts the numerical calculation of shock wave propagation, and uses numerical methods to solve the propagation law of the shock wave in the roadway space, considering factors such as the geometry of the roadway, wall reflection, and medium damping on the intensity, propagation speed, and attenuation characteristics of the shock wave. Thus, the pressure distribution and propagation range of the shock wave at different times and different spatial locations are obtained, simulating the propagation process of the underground explosion disaster in the virtual environment and obtaining the propagation simulation results.

[0080] S205: Based on the propagation simulation results, a particle system driven by a physics engine generates dynamic effects such as explosion flames and smoke diffusion. Combined with the sound wave equation, the attenuation law of the shock wave is calculated to construct multi-channel stereo sound field data.

[0081] Preferably, the physical properties of the particle system (such as particle velocity, lifespan, and color decay) are used to simulate the combustion and flickering of flames, and the diffusion and rising of smoke, with the motion laws following the basic principles of fluid mechanics. Simultaneously, the attenuation law of the shock wave in the tunnel is calculated by combining the acoustic wave equation, considering the influence of tunnel wall reflection and medium absorption on sound pressure, thereby constructing multi-channel stereo sound field data to achieve three-dimensional positioning and dynamic changes of spatial sound effects in explosion scenarios, providing a realistic auditory dimension for the virtual training environment.

[0082] S206: Calculate the regional hazard probability based on gas concentration, toxic gas diffusion range, temperature field distribution, and tunnel structural integrity in a multi-physics field coupled simulation model, classify hazard levels, and generate a spatiotemporally dynamic hazard heat map.

[0083] The formula for calculating the regional hazard probability is as follows:

[0084]

[0085] Where (x,y,z) represents spatial coordinates, and t represents time; P risk (x,y,z,t) represents the regional hazard probability; f i (x,y,z,t) represents the hazard function of the i-th physical field, i∈{1,2,3,4}, where the physical fields include methane concentration, toxic gas concentration, temperature, and structural damage; ω i (t) represents the dynamic weighting coefficient of the i-th physical field; g(t) represents the spatiotemporal decay function; f gas (x,y,z,t) represents the gas concentration function; C(x,y,z,t) represents the gas concentration at time t at coordinates (x,y,z); C exp Indicates the explosive limit concentration; C low α represents the lower threshold; M represents the sensitivity coefficient; f represents the indicator function; tox (x,y,z,t) represents the toxic gas function; D(x,y,z,t) represents the comprehensive toxicity index at time t at coordinates (x,y,z); D th Represents the danger threshold; β represents the slope of the control probability mutation; f temp (x,y,z,t) represents the temperature field function; T(x,y,z,t) represents the temperature at time t at coordinates (x,y,z); T0 represents the ambient temperature reference value; T eirt Indicates the critical temperature; f struct(x, y, z, t) represents a structure damage function; σ(x, y, z, t) represents a structure stress at the coordinate (x, y, z) at the time t; σ y represents a yield strength; γ represents a material sensitivity coefficient.

[0086] Preferably, wherein t0represents a time when a disaster occurs; t max represents a duration; d(x, y, z) represents a distance to a disaster source; d0represents a characteristic distance, which is a key parameter for measuring a spatial attenuation speed of disaster influence, when the distance d(x, y, z) to the disaster source is equal to d0, a spatial attenuation factor of the disaster influence is e -1 (about 0.368), that is, the influence intensity of the disaster at the distance is 36.8% of that at the disaster source, the parameter can be preset according to the geometric structure of the mine roadway, the medium propagation characteristics, etc., and is used to quantify the attenuation law of the disaster risk with the spatial diffusion, so that the dangerous probability calculation is more in line with the actual propagation characteristics of the underground disaster.

[0087] Preferably, wherein λ i represents an influence factor of the i th physical field.

[0088] The gas concentration function f gas (x, y, z, t) is characterized by taking the current gas concentration C(x, y, z, t) and the explosion limit concentration C exp , multiplying the indicator function , and taking the α th power of the ratio, wherein the indicator function is used to judge whether the gas concentration exceeds the lower threshold C low .

[0089] The toxic gas function f tox (x, y, z, t) is calculated by comparing the comprehensive toxicity index D(x, y, z, t) and the dangerous threshold D th , and combining the parameter β for controlling the probability mutation slope.

[0090] The temperature field function f temp (x, y, z, t) is calculated by calculating the difference between the current temperature T(x, y, z, t) and the environmental temperature reference value T0, and comparing the difference with the difference between the critical temperature T eirt and the environmental temperature reference value T0, taking the minimum value of the ratio and 1, so as to reflect the influence of the temperature field change on the dangerous probability.

[0091] The structure damage function f struct (x, y, z, t) is calculated by comparing the structure stress σ(x, y, z, t) and the yield strength σ yThe ratio is adjusted by a material sensitivity coefficient γ, and the structure damage degree is obtained by exponential operation, thereby reflecting the influence of the roadway structure integrity on the dangerous probability.

[0092] The embodiment integrates the key risk factors of multiple physical fields in a mine underground, and realizes accurate assessment of regional dangerous probability by quantifying the dangerous degree of each physical field and its spatio-temporal variation characteristics. Compared with single factor or empirical judgment, the formula can dynamically reflect the risk evolution process under complex scenarios such as gas leakage, equipment failure and environmental mutation, provide data support for dangerous heat map generation, assist training personnel to quickly identify high-risk areas, and also provide scientific risk assessment basis for mine safety management, and improve the pertinence and practicality of training content.

[0093] S207: According to the stereo sound field data and the dangerous heat map, a virtual training environment is generated by using real-time rendering and dynamic labeling technology.

[0094] The embodiment can simulate the occurrence and propagation process of underground disasters by multi-physical field coupling modeling and algorithm optimization; combined with dynamic rendering and risk assessment technology, the simulation results are converted into visual and interactive virtual training environment. The dangerous heat map quantifies the dangerous level in real time based on multi-dimensional parameters such as gas concentration, toxic gas diffusion, temperature field and structure damage, and presents the underground disaster diffusion path and high-risk area in a visual way. Training personnel can master the risk distribution law under different disaster scenarios and strengthen the risk identification ability; at the same time, the dynamically updated dangerous level can simulate the disaster evolution process, assist training personnel to learn emergency escape path planning and decision logic, and change the safety training from static knowledge infusion to dynamic risk response operation, which significantly improves the authenticity and practicality of training content, provides quantifiable and interactive training tools for mine workers' emergency ability training, and effectively reduces the actual operation risk. Compared with the traditional training method, the module can accurately reproduce complex disaster scenarios, provide an immersive learning experience for training personnel, effectively improve the authenticity and effectiveness of mine safety training, and help improve the disaster response ability of training personnel.

[0095] S3: Obtain operation behavior data of the training personnel in the virtual training environment, and evaluate the operation behavior data to obtain a behavior evaluation result;

[0096] S301: Collect the body movement, gaze trajectory and equipment operation data of the training personnel in the virtual training environment to obtain an original behavior data set;

[0097] Specifically, the motion capture device, eye tracker and interaction sensor are used to collect the body movements, eye movement trajectories and device operation data of the training personnel in the virtual environment, forming the original behavior data set. The body movement data is the movement information of each part of the body of the training personnel in the virtual training environment, such as the position of the joints of the limbs, the change of the posture (walking, climbing, gesture operation, etc.), and the movement trajectory of the limbs when interacting with the virtual device. The eye movement trajectory data is the moving path of the line of sight of the training personnel in the virtual scene obtained by eye tracking technology, including the position of the fixation point, the fixation time, the scanning frequency and the visual attention mode to the dangerous area or the key device. The device operation data is the operation behavior record of the training personnel on the virtual underground device (such as mining machinery, ventilation device, emergency rescue tool, etc.), including the operation steps, the operation sequence, the button pressing time, the device parameter adjustment amplitude and other interaction data.

[0098] S302: Data cleaning is performed on the original behavior data set, abnormal values are removed by a noise reduction algorithm, and data standardization technology is used to unify the data format, to obtain the preprocessed behavior data;

[0099] S303: Based on the preprocessed behavior data, the spatio-temporal features of the action sequence are extracted by using the spatio-temporal graph convolution network, and the key operation behavior features are strengthened by using the attention mechanism, to obtain the behavior feature vector;

[0100] S304: The behavior feature vector is subjected to an evaluation algorithm based on the fusion of rule engine and machine learning, and the safety operation specification and emergency handling process are compared to calculate the operation compliance, response time and other behavior evaluation indexes;

[0101] Specifically, the rule engine constructs a rule library based on the pre-set safety operation specification and emergency handling process (such as device operation step standard, emergency avoidance process rule, etc.). When evaluating the behavior feature vector of the training personnel, the rule engine matches the rules in the rule library to logically judge the compliance of the operation behavior (such as whether the operation steps are complete and the sequence is correct), and combines the quantitative results output by the machine learning algorithm to jointly calculate the behavior evaluation indexes, realizing the standardized evaluation of the operation behavior of the training personnel.

[0102] The behavior evaluation indexes include:

[0103] (1) Based on the operation behavior data, the basic operation compliance indexes including operation step integrity, operation sequence correctness and device use proficiency are obtained by using the spatio-temporal alignment algorithm.

[0104] (2) The behavior data after the emergency is triggered is subjected to causal inference to obtain the emergency response efficiency indexes including response delay time, decision accuracy and avoidance path rationality.

[0105] (3) Combine eye tracking data with physiological signals to calculate attention dispersion and risk identification speed through a multi-modal fusion neural network, and obtain cognitive load and situational awareness indicators based on a working memory capacity model.

[0106] The working memory capacity model is used to evaluate the cognitive load and situational awareness of the training personnel. Based on the working memory theory, it quantifies the ability of the training personnel to temporarily store, process and integrate information in the virtual environment into computable indicators. The model combines eye tracking data (such as gaze point trajectory, information retrieval path) and physiological signals (such as heart rate variability, skin electrical response) to estimate the resource occupancy rate of working memory through Bayesian inference algorithm, and then obtain the cognitive load indicator. At the same time, by analyzing the correlation between the scanning mode of eye movement data on dangerous areas and decision-making behavior, the perception and understanding of environmental information by the training personnel are evaluated, and the situational awareness indicator is determined.

[0107] After obtaining the eye tracking data (such as gaze point position, saccade trajectory) and physiological signal data (heart rate, skin electrical response) of the training personnel, first, the two types of data are time-synchronized and normalized for preprocessing to ensure data consistency and usability. Then, the processed data is input into a multi-modal fusion neural network. This network extracts visual attention features (such as gaze duration, interest area dwell frequency) from eye movement data and stress response features (such as heart rate variability, skin electrical fluctuation amplitude) from physiological signals through parallel sub-networks, and dynamically weights and fuses the two types of features through an attention mechanism, thereby calculating attention dispersion, a quantitative indicator reflecting the degree of concentration of the training personnel during operation, and risk identification speed, an indicator measuring the efficiency of discovering potential risks. At the same time, based on the working memory capacity model, combining the information retrieval path in eye movement data and the cognitive load representation of physiological signals, the resource occupancy rate of working memory is estimated through Bayesian inference algorithm, and then the cognitive load indicator is obtained. By analyzing the correlation between the scanning mode of eye movement data on dangerous areas and decision-making behavior, the perception and understanding of environmental information by the training personnel are evaluated, and the situational awareness indicator is finally determined.

[0108] This embodiment realizes the comprehensive quantification of the operation ability of the training personnel through a multi-dimensional data acquisition and analysis method. The basic operation compliance indicator ensures that the training personnel master the basic skills; the emergency response efficiency indicator tests their response ability in emergency situations; and the cognitive load and situational awareness indicators in-depth evaluate their psychological and cognitive state. Compared with single-dimensional evaluation, this system can more objectively and meticulously reflect the comprehensive ability of the training personnel, providing a basis for targeted training and effectively improving the quality and efficiency of mine safety training.

[0109] S305: According to the behavior evaluation indicators, use clustering algorithm to classify the behavior patterns of the training personnel, combine with analytic hierarchy process to determine the weight of each indicator, and construct the behavior evaluation result.

[0110] This embodiment realizes the whole-chain automatic processing from data collection to result output. Through multi-source data fusion and algorithm coordination, the behavior details of training personnel can be accurately captured, and the operation standardization and emergency disposal ability can be objectively quantified. Compared with the traditional manual evaluation method, this method effectively reduces subjective errors, improves evaluation efficiency and accuracy; based on clustering and weight analysis, the evaluation results can clearly present the individual behavior differences, provide data support for personalized training scheme, and help to improve the pertinence and effectiveness of mine safety training.

[0111] S4: based on the behavior evaluation results and environmental parameter change data, using machine learning to build and train the evaluation model, combining different mine geological conditions and mining process characteristics, generating customized training content;

[0112] S401: based on the behavior evaluation index and environmental parameter change data, using feature engineering technology to filter and reduce the dimension of the data, using principal component analysis algorithm to obtain high correlation training data set;

[0113] S402: use the combination model of random forest and gradient boosting tree as the initial evaluation model, and train it with the training data set;

[0114] S403: according to different mine geological conditions and mining process characteristics, use knowledge graph technology to extract key features, and combine adaptive weight adjustment algorithm to dynamically optimize the scoring weight of the initial evaluation model;

[0115] Specifically, first, a knowledge graph covering the knowledge in the fields of different mine geological conditions (such as rock hardness, fault distribution, hydrological characteristics), mining process (such as fully mechanized mining process, room and pillar mining) and other fields is constructed. Through knowledge extraction technology, entities (such as "shale layer", "hydraulic support") and entity relationships (such as "mining process-applicable geological conditions") are extracted from structured and unstructured data such as geological exploration reports and mining technology manuals, forming a knowledge network. On this basis, the node and relationship in the knowledge graph are mapped into low-dimensional vectors using graph embedding algorithm, and the key features related to training needs are extracted, such as equipment operation specifications under specific geological conditions, risk nodes of complex process, etc.

[0116] Then, the extracted key features are input into a self-adaptive weight adjustment algorithm. The algorithm dynamically calculates the weight coefficient of each feature based on the contribution of different features in the training data to the evaluation results. Specifically, the gradient descent method is used to minimize the error between the model prediction results and the actual training needs, and the weight is iteratively updated. For example, in a mine area with complex geological structure, the algorithm will automatically increase the weight of related features such as "geological condition cognition" and "emergency support operation"; in a mine area using new technology, the importance of features such as "equipment commissioning process" and "process parameter monitoring" is strengthened, thereby realizing the dynamic optimization of the scoring weight of the individual evaluation model, making the model more suitable for the actual training needs of the specific mine area.

[0117] S404: The optimized evaluation model is simulated by a reinforcement learning algorithm to obtain feedback in different training scenarios, and parameters are adjusted to obtain a refined evaluation model;

[0118] S405: Based on the output results of the refined evaluation model, natural language generation technology and interactive scene construction algorithm are used to generate customized training content including theoretical explanation, practical simulation and emergency drill.

[0119] The embodiment can deeply mine the correlation features of the training personnel behavior and the environmental parameters, accurately evaluate the individual training needs, and optimize the evaluation model combined with the actual characteristics of the mine area to ensure that the training content meets the real needs of different operation scenarios. Compared with the traditional fixed training mode, the adaptability and effectiveness of the training content can be significantly improved, helping the training personnel to efficiently master the corresponding skills, while reducing the repeated training cost of the enterprise, and providing a scientific and efficient solution for mine safety training.

[0120] Embodiment two:

[0121] As shown in Figure 1 The present application provides a mine safety training system based on digital twinning and virtual reality, comprising: a data acquisition modeling module, a disaster simulation generation module, a behavior acquisition evaluation module and a training content generation module;

[0122] The data acquisition modeling module is used to obtain laser scanning point cloud data and geological exploration data of a mine underground tunnel, and a high-precision virtual underground environment model is constructed based on the data using a three-dimensional reconstruction algorithm, and the virtual underground environment model is transmitted to the disaster simulation generation module;

[0123] The disaster simulation generation module is used to trigger disaster simulation based on the virtual underground environment model combined with an explosion condition judgment algorithm, generate a virtual training environment using real-time rendering and dynamic labeling technology, and transmit the virtual training environment to the behavior acquisition evaluation module;

[0124] The behavior collection and evaluation module is configured to acquire operation behavior data of the training personnel in the virtual training environment, evaluate the operation behavior data by using a behavior analysis algorithm to obtain a behavior evaluation result, and transmit the behavior evaluation result to the training content generation module.

[0125] The training content generation module is configured to train a personalized evaluation model by using a machine learning algorithm based on the behavior evaluation result and the environment parameter change data, generate customized training content in combination with different geological conditions of mining areas and mining process characteristics.

[0126] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A mine safety training method based on digital twins and virtual reality, characterized in that, include: S1: Acquire laser scanning point cloud and geological exploration data of underground mine roadways, and construct a high-precision virtual underground environment model based on the data using a three-dimensional reconstruction algorithm; S2: Based on the aforementioned virtual underground environment model and combined with explosion condition judgment, a disaster simulation is triggered, and a virtual training environment is generated using real-time rendering and dynamic annotation technology; specifically including: S201: Based on the virtual underground environment model, construct a multi-physics field coupled simulation model that includes airflow, methane concentration, toxic gas concentration, and temperature in the tunnel; S202: The gas flow field distribution in the tunnel is solved by computational fluid dynamics algorithm, and the spatiotemporal evolution of gas concentration is calculated by combining the reaction kinetics model. S203: Apply dynamic boundary conditions to the multiphysics coupled simulation model, combine the spatiotemporal evolution of gas concentration, and use machine learning algorithms to optimize the explosion conditions; S204: Real-time monitoring of environmental parameters in the multiphysics coupled simulation model; when explosion conditions are met, triggering disaster simulation, initiating numerical calculation of shock wave propagation, and obtaining propagation simulation results; S205: Based on the propagation simulation results, a particle system driven by a physics engine is used to generate dynamic effects of explosion flames and smoke diffusion. Combined with the sound wave equation, the attenuation law of shock waves is calculated to construct multi-channel stereo sound field data. S206: Calculate the regional hazard probability based on gas concentration, toxic gas diffusion range, temperature field distribution, and tunnel structural integrity in a multi-physics field coupled simulation model, classify hazard levels, and generate a spatiotemporally dynamic hazard heat map. The formula for calculating the regional hazard probability is as follows: Where (x,y,z) represents spatial coordinates, and t represents time; P risk (x,y,z,t) represents the regional hazard probability; f i (x,y,z,t) represents the hazard function of the i-th physical field, i∈{1,2,3,4}, where the physical fields include methane concentration, toxic gas concentration, temperature, and structural damage; ω i (t) represents the dynamic weighting coefficient of the i-th physical field; g(t) represents the spatiotemporal decay function; f gas (x,y,z,t) represents the gas concentration function; C(x,y,z,t) represents the gas concentration at time t at coordinates (x,y,z); C exp Indicates the explosive limit concentration; C low α represents the lower threshold; M represents the sensitivity coefficient; f represents the indicator function; tox (x,y,z,t) represents the toxic gas function; D(x,y,z,t) represents the comprehensive toxicity index at time t at coordinates (x,y,z); D th Represents the danger threshold; β represents the slope of the control probability mutation; f temp (x,y,z,t) represents the temperature field function; T(x,y,z,t) represents the temperature at time t at coordinates (x,y,z); T0 represents the ambient temperature reference value; T eirt Indicates the critical temperature; f struct (x,y,z,t) represents the structural damage function; σ(x,y,z,t) represents the structural stress at time t at coordinates (x,y,z); σ y γ represents the yield strength; γ represents the material sensitivity coefficient. S207: Based on stereo sound field data and hazard heat map, a virtual training environment is generated using real-time rendering and dynamic annotation technology; S3: Obtain the operational behavior data of trainees in the virtual training environment, evaluate the operational behavior data, and obtain the behavior evaluation results; S4: Based on the behavioral assessment results and environmental parameter change data, use machine learning to build and train an assessment model, and combine the geological conditions and mining technology characteristics of different mining areas to generate customized training content.

2. The method according to claim 1, characterized in that, The spatiotemporal decay function g(t) is expressed as follows: Where t0 represents the time when the disaster occurred; t max The distance to the disaster source is represented by d(x,y,z); d(x,y,z) represents the duration; d0 represents the characteristic distance, a key parameter used to measure the rate attenuation of disaster impact with spatial distance. When the distance to the disaster source d(x,y,z) equals d0, the spatial attenuation factor of the disaster impact is... The value is e -1 .

3. The method according to claim 2, characterized in that, The dynamic weighting coefficient ω of the i-th physical field i (t) is represented as follows: Where, λ i This represents the influence factor of the i-th physical field.

4. The method according to claim 1, characterized in that, S1 specifically includes: S101: Acquire laser scanning point cloud and geological exploration data of underground mine roadways, remove redundancy and noise from the data, unify the coordinate system through data registration technology, and obtain standardized preprocessed data corresponding to the real mine; S102: The standardized preprocessed data is discretized using a voxelization algorithm, and the region is divided using spatial partitioning techniques to obtain a structured data grid. S103: Based on the structured data grid, a basic three-dimensional model framework is generated using a point cloud surface reconstruction algorithm. The various regions are then integrated using stitching technology to obtain a preliminary three-dimensional environmental model. S104: Apply texture mapping algorithm to the surface of the preliminary 3D environmental model to give it material texture, and combine detail enhancement technology and micro-terrain optimization technology to generate an underground environment model; S105: Based on the downhole environment model, spatial annotation technology is used to determine the spatial location of the equipment, and the training information is semantically associated with the spatial location of the equipment to obtain a virtual downhole environment model.

5. The method according to claim 1, characterized in that, S3 specifically includes: S301: Collect data on trainees' body movements, eye movements, and device operation in the virtual training environment to obtain the raw behavior dataset; S302: Clean the original behavioral dataset, remove outliers using noise reduction algorithms, and unify the data format using data standardization techniques to obtain preprocessed behavioral data; S303: Based on the preprocessed behavioral data, a spatiotemporal graph convolutional network is used to extract the spatiotemporal features of the action sequence, and an attention mechanism is combined to enhance the key operational behavioral features to obtain a behavioral feature vector. S304: Apply an evaluation algorithm based on the fusion of rule engine and machine learning to the behavioral feature vector, and calculate the behavioral evaluation indicators of operational compliance and response time by comparing them with the preset safety operation specifications and emergency handling procedures. S305: Based on the behavioral assessment indicators, use clustering algorithms to classify the behavioral patterns of trainees, combine the analytic hierarchy process to determine the weight of each indicator, and construct the behavioral assessment results.

6. The method according to claim 1, characterized in that, S4 specifically includes: S401: Based on behavioral assessment indicators and environmental parameter change data, feature engineering techniques are used to filter and reduce the dimensionality of the data, and principal component analysis algorithm is combined to obtain a highly correlated training dataset. S402: Use a combination of random forest and gradient boosting tree as the initial evaluation model and train it with the training dataset; S403: Based on the geological conditions and mining technology characteristics of different mining areas, key features are extracted using knowledge graph technology, and the scoring weights of the initial evaluation model are dynamically optimized using an adaptive weight adjustment algorithm. S404: The optimized evaluation model is simulated using reinforcement learning algorithms to simulate feedback under different training scenarios, and the parameters are adjusted to obtain a more accurate evaluation model. S405: Based on the output of the precise assessment model, natural language generation technology and interactive scenario construction algorithm are used to generate customized training content including theoretical explanations, practical simulations and emergency drills.

7. A mine safety training system based on digital twins and virtual reality, characterized in that, The system, which operates the method according to any one of claims 1-6, comprises: a data acquisition and modeling module, a disaster simulation generation module, a behavior acquisition and evaluation module, and a training content generation module; The data acquisition and modeling module is used to acquire laser scanning point cloud data and geological exploration data of underground mine roadways. Based on the data, a high-precision virtual underground environment model is constructed using a three-dimensional reconstruction algorithm, and the virtual underground environment model is transmitted to the disaster simulation generation module. The disaster simulation generation module is used to trigger disaster simulation based on the virtual underground environment model and the explosion condition judgment algorithm, and to generate a virtual training environment using real-time rendering and dynamic annotation technology, and then transmit the virtual training environment to the behavior collection and evaluation module. The behavior collection and evaluation module is used to acquire the operational behavior data of trainees in the virtual training environment, evaluate the operational behavior data using behavior analysis algorithms, obtain the behavior evaluation results, and transmit the behavior evaluation results to the training content generation module. The training content generation module is used to train a personalized assessment model based on the behavioral assessment results and environmental parameter change data, and generate customized training content by combining the geological conditions and mining technology characteristics of different mining areas.

Images