Mine potential safety hazard intelligent early warning prediction method and system based on deep learning

Abstract

The invention provides a mine potential safety hazard intelligent early warning and prediction method and system based on deep learning, and relates to the field of mine safety early warning, and the method comprises the steps: obtaining monitoring data, and recognizing the type and position of a potential hazard based on a deep neural network; determining a propagation path according to the roadway communication relation and the airflow direction; calculating a future hidden danger strength value of each node; predicting the position of a person and calculating an accumulated exposure identification early warning moment; and the air volume value is reversely solved based on the early warning moment, the delayed early warning moment is obtained, an evacuation path is planned, and an instruction is issued. According to the invention, accurate prediction and active prevention and control of the mine potential safety hazard are realized, and the early warning accuracy and the emergency disposal efficiency are improved.

Description

Technical Field

[0001] This invention relates to mine safety early warning technology, and more particularly to an intelligent early warning and prediction method and system for mine safety hazards based on deep learning. Background Technology

[0002] Mine safety has always been a key concern in the coal industry. The complex working environment within mines easily generates various safety hazards, such as excessive gas levels, coal dust explosions, and fires. Traditional mine safety monitoring mainly relies on sensor networks to collect environmental parameters, issuing alarms based on simple thresholds, and handling safety accidents using human experience. With the development of IoT and AI technologies, intelligent mine safety early warning systems have gradually become a research hotspot. These systems can achieve early identification and prediction of safety hazards through the analysis of mine environmental data.

[0003] Traditional mine safety monitoring systems mostly rely on single-parameter threshold judgments, lacking the ability to comprehensively analyze multi-source heterogeneous data. This makes them ineffective in identifying potential hazard types and accurately locating hazard locations in complex environments, resulting in insufficient early warning accuracy. Existing technologies lack the ability to dynamically simulate the diffusion process of hazardous factors within mines, cannot predict hazard propagation by combining the actual ventilation network structure and airflow direction, and struggle to accurately assess the degree of danger changing over time at different spatial locations, thus failing to provide timely and effective early warnings for personnel safety. Current safety early warning systems often only focus on hazard detection and alarms, lacking the ability to dynamically track and predict personnel location and movement trajectories. They cannot provide personalized safety warnings based on the cumulative effects of personnel exposure risks, and also lack the ability to automatically generate ventilation control and personnel evacuation plans based on warning results, resulting in insufficient intelligence in emergency response. Summary of the Invention

[0004] This invention provides a method and system for intelligent early warning and prediction of mine safety hazards based on deep learning, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a deep learning-based intelligent early warning and prediction method for mine safety hazards, comprising: Acquire monitoring data of the mine production environment, analyze the monitoring data based on deep neural networks, and identify the types and locations of potential hazards; Based on the spatial connectivity of mine roadways and the direction of ventilation airflow, the airflow propagation path starting from the location of the hidden danger is determined, and the wind speed value and node spacing of each spatial node on the airflow propagation path are obtained to obtain the airflow propagation topology. The diffusion rate corresponding to the type of hazard is progressively propagated and calculated with the wind speed value in the airflow propagation topology to obtain the hazard intensity value of each spatial node in the future time sequence. Based on the historical movement trajectory of personnel, the spatial location of personnel in the future time sequence is predicted, the cumulative exposure is calculated according to the dwell time of personnel at each location and the hazard intensity value of the corresponding node, and the warning time is identified. Using the warning time as the constraint target, the required increase in air volume for ventilation equipment along the airflow propagation path is calculated in reverse. The wind speed increment corresponding to the air volume value is substituted into the progressive propagation calculation to obtain the delayed warning time. Based on the time interval between the delayed warning time and the current time, personnel evacuation routes are planned, and ventilation control instructions and personnel evacuation instructions are issued.

[0006] Acquire monitoring data of the mine production environment, analyze the monitoring data based on a deep neural network, and identify the types and locations of potential hazards, including: Collect gas concentration, temperature, and pressure data in the mine production environment, record the data collection time and sensor spatial coordinates, and generate monitoring data with spatiotemporal markers; Abnormal values ​​are removed from the monitoring data, and the data is aligned according to the time of collection. The aligned monitoring data is then normalized by interval transformation to generate standardized monitoring data. Standardized monitoring data is divided into time windows at fixed intervals. Within the time window, convolution operations are performed to extract local features, and nonlinear transformations are applied to the local features to generate feature vectors. A deep neural network is constructed, which includes an input layer, multiple hidden layers, and an output layer. The feature vector is input into the input layer, and through nonlinear transformation and weight transfer in the hidden layer, the probability distribution of the hazard type is calculated in the output layer. The hazard type and hazard severity value are determined based on the probability distribution. The distance matrix is ​​calculated based on the sensor's spatial coordinates. The hazard severity value is converted into a spatial weight. Interpolation calculation is then performed using the distance matrix to determine the spatial location of the hazard.

[0007] Based on the spatial connectivity and ventilation airflow direction of the mine roadways, the airflow propagation path starting from the location of the potential hazard is determined. The wind speed values ​​and node spacing of each spatial node along the airflow propagation path are obtained, resulting in the airflow propagation topology, including: Collect spatial coordinate points and cross-sectional contour points of the tunnel, calculate the curvature value between the spatial coordinate points, identify the inflection point position based on the curvature value, use piecewise spline interpolation to connect the spatial coordinate points to generate the tunnel center curve, generate the cross-sectional boundary based on the cross-sectional contour points, and combine the tunnel center curve and the cross-sectional boundary to form a spatial connectivity relationship; Collect the power and resistance values ​​of the ventilation equipment, establish an energy transfer equation based on the spatial connectivity, include the surface roughness coefficient and inflection point position in the loss term, solve the energy transfer equation to obtain the velocity field, and determine the ventilation airflow direction based on the velocity field; Starting from the location of the potential hazard, a propagation path map is constructed along the direction of the ventilation airflow. Based on the pressure difference and velocity difference between adjacent spatial nodes in the propagation path map, the propagation probability is calculated to determine the airflow propagation path. The wind speed values ​​of each spatial node on the airflow propagation path are obtained based on the velocity field, the node spacing between spatial nodes is calculated based on the spatial coordinate points, and the flow rate between spatial nodes is calculated in combination with the cross-sectional boundary. The airflow propagation path and the flow rate between spatial nodes are combined to generate an airflow propagation topology.

[0008] By progressively propagating the diffusion rate corresponding to the hazard type to the wind speed value in the airflow propagation topology, the hazard intensity value of each spatial node in the future time sequence is obtained, including: The diffusion rate of each spatial node at the current moment is calculated based on the diffusion rate corresponding to the type of hazard and the wind speed value in the airflow propagation topology, and a progressive propagation sequence of diffusion amount is constructed. Based on the airflow propagation topology, the connection path and wind direction change path between spatial nodes are determined, the diffusion attenuation and wind speed attenuation between spatial nodes are calculated, the diffusion attenuation and wind speed attenuation are applied to the wind speed value to obtain the corrected wind speed value, and the corrected wind speed value is combined with the diffusion amount progressive propagation sequence to form the hazard propagation parameters. The hazard propagation parameters are transmitted point by point according to the connection sequence of spatial nodes in the airflow propagation topology. The amplitude of the diffusion change at the spatial node is compared with the preset amplitude threshold. The calculation time step is determined based on the comparison result, and the hazard diffusion distribution at the spatial node at each time step is obtained. The spatial node hazard diffusion distribution is unfolded according to the time series. The diffusion change value of adjacent time points and the diffusion accumulation value within the time period are calculated. The initial hazard intensity value is generated by combining them according to the preset ratio. Based on the intensity difference between adjacent time points and the intensity difference between adjacent spatial nodes, outliers are removed and missing values ​​are supplemented to obtain the hazard intensity value of each spatial node in the future time series.

[0009] Based on historical movement trajectories, the spatial location of individuals in future time sequences is predicted. Cumulative exposure is calculated based on the time individuals spend at each location and the corresponding hazard intensity value at each node. Warning times are identified, including: Collect personnel location tag data to obtain personnel identification number, timestamp and spatial coordinates, and construct a sequence of personnel's historical movement trajectory based on the timestamp order; Calculate the rate of change of acceleration and the rate of change of direction of motion between adjacent timestamps in the historical movement trajectory sequence of personnel, and combine the rate of change of acceleration and the rate of change of direction of motion to form a motion feature sequence; Based on motion feature sequences, the location of dwell points and transition points in the historical movement trajectory sequence of personnel are identified, the dwell time at the dwell point location and the transit time at the transition point location are calculated, and a spatiotemporal distribution sequence is generated. Based on the spatiotemporal distribution sequence, predict the distribution of dwell points and transition points of people in the future time sequence, map the distribution of dwell points and transition points to spatial node locations, and calculate the dwell time at spatial nodes; The exposure intensity is calculated based on the dwell time of the spatial node and the corresponding hazard intensity value. The exposure amount per unit time is calculated based on the exposure intensity according to the time dimension. The cumulative exposure amount is obtained by summing the exposure amounts per unit time. The abrupt change point of the cumulative exposure amount is identified by calculating the rate of change of the cumulative exposure amount as the warning moment.

[0010] Using the warning time as the constraint objective, the required increase in air volume for ventilation equipment along the airflow propagation path is calculated in reverse. The corresponding wind speed increment is then substituted into the progressive propagation calculation to obtain the delayed warning time, which includes: Using the warning time as a time constraint boundary condition, the upper limit of the hazard intensity at each spatial node at the warning time is inferred from the cumulative exposure threshold. The hazard intensity difference that needs to be reduced at each spatial node is calculated by tracing back from the downstream node to the upstream node along the airflow propagation path. Based on the coupling relationship between the difference in hazard intensity and wind speed, a set of air volume optimization equations is established. The node spacing in the airflow propagation topology is used as a constraint parameter. The air volume optimization equations are solved to obtain the air volume increase required for each ventilation device on the airflow propagation path. The wind speed increment of each spatial node is calculated based on the required increase in air volume of the ventilation equipment. The wind speed increment is then superimposed with the original wind speed value in the airflow propagation topology to obtain the adjusted wind speed value. Substitute the adjusted wind speed value into the progressive propagation calculation of diffusion rate and wind speed value to recalculate the adjusted hazard intensity value of each spatial node in the future time sequence; Based on the spatial location of personnel in the future time sequence and the intensity value of the hazard after the control, the cumulative exposure after the control is recalculated, and the moment when the cumulative exposure after the control reaches the cumulative exposure threshold is identified as the delayed warning moment.

[0011] Based on the time interval between the delayed warning time and the current time, evacuation routes are planned, and ventilation control instructions and personnel evacuation instructions are issued, including: The time difference between the delayed warning time and the current time is calculated as the available evacuation time. Based on the spatial connectivity of the mine roadways, a candidate set of evacuation routes is constructed from the current spatial location of personnel. Based on the personnel movement speed and candidate path length, time-feasible paths with an estimated evacuation time less than the available evacuation time are selected. The hazard intensity value of each spatial node on the time-feasible path after adjustment is obtained. The cumulative exposure of personnel moving along the time-feasible path is calculated. The time-feasible path with the smallest cumulative exposure is selected as the personnel evacuation path. Based on the required increase in air volume for the ventilation equipment, a ventilation control command containing the equipment identification number and the target air volume adjustment value is generated and sent to the control unit. The control unit returns a feedback message indicating that the adjustment is complete. After receiving the feedback message, a personnel evacuation command containing the personnel identification number and the coordinate sequence of the evacuation path is generated based on the personnel evacuation path.

[0012] A second aspect of this invention provides a deep learning-based intelligent early warning and prediction system for mine safety hazards, comprising: The first unit is used to acquire monitoring data of the mine production environment, analyze the monitoring data based on a deep neural network, and identify the types and locations of potential hazards. The second unit is used to determine the airflow propagation path from the location of the hidden danger based on the spatial connectivity of the mine roadway and the direction of ventilation airflow, obtain the wind speed value and node spacing of each spatial node on the airflow propagation path, and obtain the airflow propagation topology. The third unit is used to progressively propagate the diffusion rate corresponding to the type of hazard to the wind speed value in the airflow propagation topology, and obtain the hazard intensity value of each spatial node in the future time sequence. The fourth unit is used to predict the spatial location of people in the future time sequence based on their historical movement trajectories, calculate the cumulative exposure based on the time people stay at each location and the hazard intensity value of the corresponding node, and identify the warning time. The fifth unit is used to calculate the required increase in air volume for ventilation equipment along the airflow propagation path with the warning time as the constraint target. The wind speed increment corresponding to the air volume value is substituted into the progressive propagation calculation to obtain the delayed warning time. Based on the time interval between the delayed warning time and the current time, the personnel evacuation path is planned, and ventilation control instructions and personnel evacuation instructions are issued.

[0013] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] In this embodiment, deep neural network analysis of mine monitoring data enables timely and accurate identification of hazard types and locations, significantly improving the accuracy and real-time performance of hazard identification compared to traditional methods. By constructing an airflow propagation topology based on the spatial connectivity of mine roadways and the direction of ventilation airflow, dynamic prediction of hazard diffusion within the mine space is achieved, overcoming the limitations of traditional static monitoring. Progressive propagation calculations of hazard diffusion rate and wind speed values ​​within the airflow propagation topology make the prediction of hazard intensity values ​​at future spatial nodes more accurate and reliable. By predicting future personnel locations and calculating cumulative exposure based on hazard intensity, quantitative assessment of personnel safety risks and precise identification of warning times are achieved. Based on warning time constraints, the system can automatically calculate the required increase in airflow for ventilation equipment, perform ventilation control, and plan personnel evacuation routes, realizing a shift from passive emergency response to proactive prevention and greatly improving the scientific nature and effectiveness of mine safety management. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the intelligent early warning and prediction method for mine safety hazards based on deep learning, as described in an embodiment of the present invention. Figure 2 This is a flowchart of the mine airflow propagation topology construction method according to an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0019] Figure 1 This is a flowchart illustrating the intelligent early warning and prediction method for mine safety hazards based on deep learning, as described in an embodiment of the present invention. Figure 1 As shown, the method includes: Acquire monitoring data of the mine production environment, analyze the monitoring data based on deep neural networks, and identify the types and locations of potential hazards; Based on the spatial connectivity of mine roadways and the direction of ventilation airflow, the airflow propagation path starting from the location of the hidden danger is determined, and the wind speed value and node spacing of each spatial node on the airflow propagation path are obtained to obtain the airflow propagation topology. The diffusion rate corresponding to the type of hazard is progressively propagated and calculated with the wind speed value in the airflow propagation topology to obtain the hazard intensity value of each spatial node in the future time sequence. Based on the historical movement trajectory of personnel, the spatial location of personnel in the future time sequence is predicted, the cumulative exposure is calculated according to the dwell time of personnel at each location and the hazard intensity value of the corresponding node, and the warning time is identified. Using the warning time as the constraint target, the required increase in air volume for ventilation equipment along the airflow propagation path is calculated in reverse. The wind speed increment corresponding to the air volume value is substituted into the progressive propagation calculation to obtain the delayed warning time. Based on the time interval between the delayed warning time and the current time, personnel evacuation routes are planned, and ventilation control instructions and personnel evacuation instructions are issued.

[0020] In one optional implementation, monitoring data of the mine production environment is acquired, and the monitoring data is analyzed based on a deep neural network to identify the type and location of potential hazards, including: Collect gas concentration, temperature, and pressure data in the mine production environment, record the data collection time and sensor spatial coordinates, and generate monitoring data with spatiotemporal markers; Abnormal values ​​are removed from the monitoring data, and the data is aligned according to the time of collection. The aligned monitoring data is then normalized by interval transformation to generate standardized monitoring data. Standardized monitoring data is divided into time windows at fixed intervals. Within the time window, convolution operations are performed to extract local features, and nonlinear transformations are applied to the local features to generate feature vectors. A deep neural network is constructed, which includes an input layer, multiple hidden layers, and an output layer. The feature vector is input into the input layer, and through nonlinear transformation and weight transfer in the hidden layer, the probability distribution of the hazard type is calculated in the output layer. The hazard type and hazard severity value are determined based on the probability distribution. The distance matrix is ​​calculated based on the sensor's spatial coordinates. The hazard severity value is converted into a spatial weight. Interpolation calculation is then performed using the distance matrix to determine the spatial location of the hazard.

[0021] Collecting gas concentration, temperature, and pressure data from the mine's production environment is fundamental for hazard identification. In practical applications, various sensor devices can be deployed, including methane, carbon monoxide, and oxygen concentration sensors, as well as temperature and pressure sensors, forming a distributed sensor network. Each sensor is equipped with a unique identifier and a high-precision positioning module, capable of recording the data acquisition time and corresponding three-dimensional spatial coordinates. The acquisition frequency can be adjusted according to the mine's safety level; typically, gas concentration data is collected every 30 seconds, temperature data every minute, and pressure data every 2 minutes. The collected raw data, marked with timestamps and spatial coordinates, forms a monitoring data sequence with spatiotemporal information, providing complete spatiotemporal information support for subsequent analysis.

[0022] The quality of monitoring data directly affects the accuracy of hazard identification, thus requiring rigorous data preprocessing. First, outlier removal is performed using the three-standard-deviation method. This involves calculating the mean and standard deviation of each data point within a specific time period, and marking and removing values ​​that deviate from the mean by more than three standard deviations. For different types of sensor data, considering the differences in acquisition frequency, time-series alignment is necessary. Using the highest-frequency data as a benchmark, low-frequency data is supplemented using linear interpolation to ensure all data have corresponding values ​​at a unified time point. After time-series alignment, normalization is performed on data of different dimensions using a maximum-minimum normalization method, mapping various data types to the [0, 1] interval. This normalization eliminates dimensional differences between different physical quantities, generating standardized monitoring data that facilitates subsequent training and application of deep learning models.

[0023] The standardized monitoring data exhibits high-dimensional temporal characteristics, requiring further processing through time windowing and feature extraction. The standardized monitoring data is divided into continuous time windows at fixed 10-minute intervals, with a 5-minute overlap between adjacent windows to enhance the model's sensitivity to temporal changes. Within each time window, a one-dimensional convolutional neural network is applied to extract local temporal features. The kernel size is set to 5, the stride to 1, and 32 different kernels are used in parallel to extract features. After the convolution operation, a non-linear transformation is performed using the ReLU activation function to enhance the model's expressive power. After convolutional feature extraction, max pooling is applied to reduce feature dimensionality while retaining salient features. The pooling window size is 2, and the stride is 2, effectively reducing the dimensionality of the feature vector. The pooled features are further transformed through a fully connected layer to generate fixed-dimensional feature vectors, which serve as the input to the deep neural network.

[0024] The constructed deep neural network employs a multilayer perceptron structure, specifically designed for hazard type identification and severity assessment. The network consists of one input layer, three hidden layers, and one output layer. The number of nodes in the input layer matches the dimension of the feature vector, the number of nodes in the hidden layers are 256, 128, and 64 respectively, and the number of nodes in the output layer equals the predefined number of hazard types. The hidden layers utilize the Leaky ReLU activation function to introduce a non-linear transformation, avoiding the "dead neuron" problem that may occur with traditional ReLU. Dropout layers are added between network layers with a random dropout rate of 0.3 to effectively prevent overfitting. The output layer uses the Softmax activation function to calculate the probability distribution of each hazard type. Model training employs the cross-entropy loss function, using the Adam optimizer for parameter updates. The initial learning rate is set to 0.001, and a learning rate decay strategy is implemented. During training, early stopping is used to prevent overfitting; training stops when the loss function value on the validation set does not decrease for five consecutive epochs. The trained deep neural network can obtain the probability distribution of hazard types from the output layer, selecting the type with the highest probability as the identification result, and simultaneously using this probability value as a quantitative indicator of hazard severity.

[0025] Determining the precise spatial location of a hazard requires spatial interpolation calculations combining sensor location information and hazard severity values. First, the distance matrix D between the sensors is calculated based on the three-dimensional coordinates of all sensors. The matrix elements D... ij Let represent the Euclidean distance between sensor i and sensor j. The hazard severity value output by the neural network is converted into a spatial weight W for the sensor location, with the weight proportional to the hazard severity value. An inverse distance weighted interpolation method is used to mesh the three-dimensional space of the mine, with a mesh size of 1 meter × 1 meter × 1 meter. For any point p in the space, its hazard severity value H(p) is calculated as follows: ; Where N is the number of sensors, W i Let d(p, i) be the hazard level value at sensor i, and d(p, i) be the distance from point p to sensor i. This method calculates the hazard level distribution across the entire mine space and visualizes the hazard distribution using isosurface methods. The hazard location is defined as the geometric center of the area where the hazard level value exceeds a preset threshold (usually 0.7), and the three-dimensional coordinates of this center point are the final determined hazard location.

[0026] This method can identify potential gas accumulation hazards and their locations 10-15 minutes in advance by using a deep learning model to comprehensively analyze historical and current data in the early stages when sensors detect slight abnormal changes in gas concentration. This is much earlier than the trigger time of traditional gas concentration alarm thresholds, providing a valuable time window for subsequent hazard spread prediction and personnel evacuation.

[0027] The identified hazard types and locations will serve as key inputs for subsequent airflow propagation path analysis, hazard intensity prediction, and personnel evacuation planning, supporting the operation of the entire mine safety hazard intelligent early warning and prediction system. Hazard location coordinates will also be used to determine the starting point of airflow propagation, while the hazard type determines the diffusion characteristic parameters of different hazardous substances. This information collectively constitutes the foundational conditions for hazard propagation simulation.

[0028] like Figure 2 The diagram shows a flowchart of the mine airflow propagation topology construction method in this embodiment.

[0029] In one optional implementation, based on the spatial connectivity of the mine roadways and the direction of ventilation airflow, the airflow propagation path originating from the location of the potential hazard is determined, and the wind speed values ​​and node spacings of each spatial node along the airflow propagation path are obtained, resulting in an airflow propagation topology including: Collect spatial coordinate points and cross-sectional contour points of the tunnel, calculate the curvature value between the spatial coordinate points, identify the inflection point position based on the curvature value, use piecewise spline interpolation to connect the spatial coordinate points to generate the tunnel center curve, generate the cross-sectional boundary based on the cross-sectional contour points, and combine the tunnel center curve and the cross-sectional boundary to form a spatial connectivity relationship; Collect the power and resistance values ​​of the ventilation equipment, establish an energy transfer equation based on the spatial connectivity, include the surface roughness coefficient and inflection point position in the loss term, solve the energy transfer equation to obtain the velocity field, and determine the ventilation airflow direction based on the velocity field; Starting from the location of the potential hazard, a propagation path map is constructed along the direction of the ventilation airflow. Based on the pressure difference and velocity difference between adjacent spatial nodes in the propagation path map, the propagation probability is calculated to determine the airflow propagation path. The wind speed values ​​of each spatial node on the airflow propagation path are obtained based on the velocity field, the node spacing between spatial nodes is calculated based on the spatial coordinate points, and the flow rate between spatial nodes is calculated in combination with the cross-sectional boundary. The airflow propagation path and the flow rate between spatial nodes are combined to generate an airflow propagation topology.

[0030] After acquiring monitoring data of the mine production environment and identifying the types and locations of potential hazards using deep neural networks, it is necessary to construct the propagation path of hazardous substances in the mine space. First, spatial coordinate points of the mine roadways are collected. These coordinate points are typically measured every five to ten meters along the roadway centerline using a 3D lidar or total station. Simultaneously, roadway cross-sectional profile points are collected, usually 16 to 32 points around the roadway cross-section at each spatial coordinate point, forming a closed profile.

[0031] The collected sequence of spatial coordinate points in the tunnel is processed to calculate the curvature value formed by three adjacent points. Specifically, for the sequence of spatial coordinate points, the curvature value of each point is calculated, which can be represented by the reciprocal of the radius of the circle formed by the three adjacent points. When the curvature value exceeds a preset threshold (usually 0.1 to 0.3), the point is marked as an inflection point. These inflection points are key locations where the tunnel turns or branches, and have a significant impact on changes in airflow direction.

[0032] These spatial coordinate points are connected using piecewise spline interpolation to generate a smooth roadway center curve. This process employs cubic spline interpolation to ensure the curve has continuous first and second derivatives at each sampling point, thus accurately reflecting the roadway's tortuosity. Based on the cross-sectional profile points, the boundary representation of the roadway's cross-section is generated. By combining the roadway center curve with the cross-sectional boundary at each location, a complete three-dimensional roadway model is constructed, thus forming a precise expression of the mine's spatial connectivity.

[0033] The power and air resistance values ​​of each ventilation device in the mine are collected. Ventilation equipment includes main fans, local fans, and dampers, and their power values ​​are typically obtained from equipment nameplates or real-time monitoring systems. Air resistance values ​​are obtained through on-site measurement or by consulting equipment specification tables. Based on the established spatial connectivity, an energy transfer equation is constructed. This equation, based on fluid mechanics principles, describes the flow behavior of ventilation airflow in the mine roadway network.

[0034] The energy transfer equation needs to consider energy losses caused by surface roughness and inflection point locations. The surface roughness of the tunnel is typically determined based on the tunnel support type; for example, the roughness coefficient for concrete support is approximately 0.012, while that for timber support is approximately 0.025. Energy losses at inflection points are proportional to the inflection angle and the square of the airflow velocity. These factors are incorporated into the loss term of the energy transfer equation, making the model more closely reflect reality.

[0035] The energy transfer equation is solved using numerical methods (such as the finite element method or the finite difference method) to obtain the velocity field distribution in the mine roadway network. The velocity field provides information on the direction and speed of airflow at various locations in the mine, providing a basis for determining the direction of ventilation airflow. Based on the vector direction of the velocity field, the airflow direction within each roadway segment is determined, thus clarifying the airflow direction diagram of the entire mine's ventilation network.

[0036] After identifying the location of the hazard, a propagation path map is constructed along the direction of the ventilation airflow. Specifically, starting from the location of the hazard as the starting node, possible propagation paths are organized into a directed graph structure along the airflow direction. For each pair of adjacent spatial nodes in the propagation path map, the propagation probability is calculated based on the pressure and velocity differences between them. By setting a probability threshold (usually 0.15), high-probability propagation paths are filtered out, thereby determining the main airflow propagation path of the hazardous substance.

[0037] Based on the solved velocity field, the wind speed values ​​of each spatial node along the airflow propagation path are extracted. Spatial nodes are typically located at intersections, inflection points, and intervals of 20 to 50 meters along the tunnel. Simultaneously, based on the previously collected spatial coordinates, the distance between adjacent spatial nodes, i.e., the node spacing, is calculated. This distance is calculated using three-dimensional Euclidean distance, taking into account the actual curvature of the tunnel path.

[0038] Based on the established tunnel cross-sectional boundary information, the flow rate between spatial nodes is calculated. The flow rate Q is equal to the product of the wind speed v and the cross-sectional area S, i.e., Q = v × S. For cases with irregular cross-sectional areas, numerical integration methods can be used for calculation. The flow rate between nodes reflects the volume of air passing through that tunnel segment per unit time and is an important parameter for assessing the intensity of hazard propagation.

[0039] Finally, the determined airflow propagation paths are combined with the calculated flow data between spatial nodes to generate an airflow propagation topology. This topology is represented as a directed weighted graph, where nodes represent spatial locations, edges represent airflow propagation paths, and the weight of each edge represents the corresponding flow value. This airflow propagation topology visually demonstrates the complete situation of airflow propagation in the mine roadway network, providing necessary input data for subsequent calculations of the hazard intensity values ​​of each spatial node at future times.

[0040] In practical applications, the airflow propagation topology is visualized as a 3D network diagram. Paths with different flow rates can be represented by different line widths or colors, facilitating mine safety management personnel's intuitive understanding of the hazard propagation situation. Furthermore, this topology supports dynamic updates; when mine ventilation system parameters change, relevant parameters can be adjusted and recalculated in a timely manner, ensuring the accuracy of the early warning system.

[0041] The airflow propagation topology constructed using the above method lays the foundation for subsequent progressive propagation calculations of the diffusion rate corresponding to the type of hazard and the wind speed value in the airflow propagation topology. This enables accurate prediction of the hazard intensity value of each spatial node in the future time sequence, achieving intelligent early warning and prediction of mine safety hazards.

[0042] In one optional implementation, the diffusion rate corresponding to the hazard type is progressively propagated and calculated with the wind speed value in the airflow propagation topology to obtain the hazard intensity value of each spatial node in the future time sequence, including: The diffusion rate of each spatial node at the current moment is calculated based on the diffusion rate corresponding to the type of hazard and the wind speed value in the airflow propagation topology, and a progressive propagation sequence of diffusion amount is constructed. Based on the airflow propagation topology, the connection path and wind direction change path between spatial nodes are determined, the diffusion attenuation and wind speed attenuation between spatial nodes are calculated, the diffusion attenuation and wind speed attenuation are applied to the wind speed value to obtain the corrected wind speed value, and the corrected wind speed value is combined with the diffusion amount progressive propagation sequence to form the hazard propagation parameters. The hazard propagation parameters are transmitted point by point according to the connection sequence of spatial nodes in the airflow propagation topology. The amplitude of the diffusion change at the spatial node is compared with the preset amplitude threshold. The calculation time step is determined based on the comparison result, and the hazard diffusion distribution at the spatial node at each time step is obtained. The spatial node hazard diffusion distribution is unfolded according to the time series. The diffusion change value of adjacent time points and the diffusion accumulation value within the time period are calculated. The initial hazard intensity value is generated by combining them according to the preset ratio. Based on the intensity difference between adjacent time points and the intensity difference between adjacent spatial nodes, outliers are removed and missing values ​​are supplemented to obtain the hazard intensity value of each spatial node in the future time series.

[0043] Diffusion rate data corresponding to different hazard types is obtained, which can be acquired through historical data statistics or professional databases. Different types of mine hazards, such as methane, dust, and toxic gases, exhibit different diffusion characteristics and rates. For example, the diffusion rate of methane gas in still air is approximately 0.1 to 0.5 m / s, while in the presence of airflow, its diffusion rate is influenced by the airflow speed. The basic diffusion rate data for each hazard type is combined with the wind speed values ​​obtained from the airflow propagation topology to calculate the current hazard diffusion amount at each spatial node. Specifically, for a given node, the current hazard diffusion amount can be expressed as a function of the inherent diffusion coefficient of the hazard type and the wind speed at that node. Considering the nonlinear relationship between the diffusion coefficient and wind speed, a suitable mathematical model can be used for description. Based on the calculated current hazard diffusion amounts at each node, a progressive propagation sequence of diffusion amount is constructed according to time series and spatial location. This sequence includes the diffusion amount distribution of the hazard source location and surrounding nodes in the initial state.

[0044] Based on the established airflow propagation topology, the connection paths and wind direction change paths between spatial nodes are determined. The airflow propagation topology includes information such as the geometric layout of the mine roadway, the spatial relationships between nodes, and the airflow direction. According to fluid mechanics principles, airflow propagation in the roadway is affected by factors such as wall friction, cross-sectional changes, and bends, leading to attenuation during diffusion. The diffusion attenuation between spatial nodes is calculated, which is related to factors such as the distance between nodes, roadway geometry, and wall roughness. Simultaneously, the wind speed attenuation between nodes is calculated, considering the influence of roadway resistance and cross-sectional changes on airflow velocity. The calculated diffusion attenuation and wind speed attenuation are applied to the original wind speed value to obtain a corrected wind speed value. The corrected wind speed value more accurately reflects the airflow propagation characteristics in the actual roadway. The corrected wind speed value is combined with the diffusion amount progressive propagation sequence constructed in the previous step to form a complete hazard propagation parameter. This parameter includes multi-dimensional data such as node location, corrected wind speed, diffusion amount, and time information.

[0045] Following the connection sequence of spatial nodes in the airflow propagation topology, the hazard propagation parameters formed in the previous step are transmitted point by point, starting from the location of the hazard source. For each spatial node, based on the parameters received from the upstream node and combined with the characteristic parameters of the node itself, the hazard diffusion state at the current moment is calculated. During the parameter transmission process, the amplitude of the diffusion change at each spatial node is compared with a preset amplitude threshold. When the amplitude of the change is greater than the threshold, it indicates that the hazard diffusion in that area is changing drastically, requiring a smaller calculation time step to improve calculation accuracy; conversely, a larger time step can be used to improve calculation efficiency. The calculation time step is dynamically adjusted based on the comparison results, optimizing the utilization of computing resources while ensuring calculation accuracy. Through iterative calculations, the hazard diffusion distribution of spatial nodes at each time step is finally obtained, which characterizes the diffusion process of hazardous materials in the mine space over time.

[0046] The spatial node hazard diffusion distribution obtained in the previous step is expanded into a time-space two-dimensional matrix, where each element represents the hazard diffusion amount at a specific spatial node at a specific time point. The diffusion amount change value between adjacent time points is calculated, reflecting the dynamic process of hazard diffusion. Simultaneously, the cumulative diffusion amount within each time period is calculated, representing the accumulated hazard exposure at a certain node over a period. The diffusion amount change value and the cumulative diffusion amount value are weighted and combined according to a preset proportional coefficient to generate initial hazard intensity values. These initial intensity values ​​may contain outliers or missing values, requiring data processing. By analyzing the distribution characteristics of intensity differences between adjacent time points and intensity differences between adjacent spatial nodes, outliers that significantly deviate from the normal range are identified and removed. For missing data, interpolation algorithms, such as linear interpolation and spline interpolation, are used to supplement the missing values. After outlier removal and missing value supplementation, the hazard intensity values ​​for each spatial node in the future time sequence are obtained. This intensity value sequence comprehensively describes the change in hazard diffusion intensity over time in the mine space, providing fundamental data for subsequent personnel exposure calculations and early warning time determination.

[0047] In practical applications, specific diffusion models can be established for different types of mine hazards. For example, for gas exceeding limits, the diffusion characteristics at different concentrations are considered; for fire hazards, the influence of temperature gradients on airflow needs to be considered. The calculation results of hazard intensity values ​​are visualized, forming a hazard diffusion cloud map that changes over time, intuitively showing the hazard diffusion range and intensity, and providing support for safety management decisions.

[0048] In one optional implementation, the spatial location of personnel in the future time sequence is predicted based on their historical movement trajectories. The cumulative exposure is calculated based on the time spent at each location and the corresponding hazard intensity value at each node. Warning times are identified, including: Collect personnel location tag data to obtain personnel identification number, timestamp and spatial coordinates, and construct a sequence of personnel's historical movement trajectory based on the timestamp order; Calculate the rate of change of acceleration and the rate of change of direction of motion between adjacent timestamps in the historical movement trajectory sequence of personnel, and combine the rate of change of acceleration and the rate of change of direction of motion to form a motion feature sequence; Based on motion feature sequences, the location of dwell points and transition points in the historical movement trajectory sequence of personnel are identified, the dwell time at the dwell point location and the transit time at the transition point location are calculated, and a spatiotemporal distribution sequence is generated. Based on the spatiotemporal distribution sequence, predict the distribution of dwell points and transition points of people in the future time sequence, map the distribution of dwell points and transition points to spatial node locations, and calculate the dwell time at spatial nodes; The exposure intensity is calculated based on the dwell time of the spatial node and the corresponding hazard intensity value. The exposure amount per unit time is calculated based on the exposure intensity according to the time dimension. The cumulative exposure amount is obtained by summing the exposure amounts per unit time. The abrupt change point of the cumulative exposure amount is identified by calculating the rate of change of the cumulative exposure amount as the warning moment.

[0049] In practice, the first step is to collect personnel location tag data using a positioning system deployed within the mine. This location tag data contains three basic elements: personnel identification number, timestamp, and spatial coordinates. The personnel identification number uniquely identifies each worker in the mine, the timestamp records the precise time of data collection, and the spatial coordinates represent the worker's specific location within the mine in three dimensions. The positioning system can be an indoor positioning network based on technologies such as ultra-wideband, RFID, or WiFi. The sampling frequency is typically set to 1-5 times per second to ensure the continuity and integrity of the trajectory data. After data cleaning to remove outliers and noise, the collected raw positioning data is arranged chronologically according to the timestamps to construct a historical movement trajectory sequence for each worker.

[0050] Based on the constructed historical movement trajectory sequence, the characteristics of a person's motion state changes between adjacent time points are calculated. Specifically, for any two adjacent time points in the trajectory sequence, the rate of change of acceleration and the rate of change of motion direction are calculated. The rate of change of acceleration is obtained by dividing the difference in acceleration between two adjacent time intervals by the time difference; the rate of change of motion direction is obtained by dividing the angle between the motion direction vectors of two adjacent time intervals by the time difference. The calculated sequences of acceleration and motion direction changes are combined to form a motion feature sequence, which reflects the characteristics of a person's motion state changes during movement, providing basic data support for subsequent behavior pattern recognition.

[0051] The motion feature sequence contains key information about personnel movement behavior. By performing pattern recognition on this sequence, different activity states of personnel in the mine can be distinguished. Based on thresholds for the rate of change of acceleration and the rate of change of movement direction, points in the personnel movement trajectory are divided into two categories: dwell points and transition points. Dwell points are locations where personnel stay and work in a certain area, characterized by small rates of change of acceleration and movement direction, with movement speed close to zero. Transition points are locations traversed by personnel when moving between different dwell points, characterized by large rates of change of acceleration or movement direction. For identified dwell points, the dwell time at that location is calculated, i.e., the time difference from entering the dwell state to leaving the state; for transition points, the time required for personnel to traverse the area is calculated. Through these calculations, a spatiotemporal distribution sequence containing the dwell / transition states of personnel at each location and their corresponding times is generated.

[0052] The spatiotemporal distribution sequence reflects the regularity of personnel activities in the mine space, and based on this sequence, the possible locations of personnel at future moments can be predicted. The prediction process employs time series analysis methods, combined with deep learning models such as recurrent neural networks or long short-term memory networks. By learning historical movement patterns of personnel, it predicts the possible locations of dwelling points and transition points for personnel within a future period. The prediction model considers factors such as personnel work shifts, work nature, and historical behavioral patterns, outputting the probability distribution of dwelling points and transition points for future time sequences. Mapping these predicted dwelling points and transition points to node locations in the mine's spatial topology yields the spatial nodes where personnel may stay in the future and the duration of their stay at each node.

[0053] After obtaining the possible future locations of personnel, the exposure intensity of personnel at each location is calculated by combining the hazard intensity values ​​of each spatial node obtained in the previous steps with the hazard intensity values ​​of the future time sequence. The formula for calculating the exposure intensity is the product of the hazard intensity value of the spatial node and the time the personnel stay at that node. Based on the exposure intensity distribution along the time dimension, the exposure amount received by personnel in each time unit is calculated, i.e., the exposure amount per unit time. As time progresses, the exposure amounts of each time unit are accumulated to obtain the cumulative exposure curve. By analyzing the changing characteristics of the cumulative exposure curve, the first derivative of the cumulative exposure, i.e., the rate of change, is calculated. When the rate of change exceeds a preset safety threshold or shows a significant increase in a short period of time, this time point is identified as the abrupt change point of the cumulative exposure, and this moment is determined as the safety warning moment.

[0054] The determination of the warning time takes into account the actual location and movement patterns of personnel in the mine, as well as the spatiotemporal characteristics of hazard spread, enabling personalized warnings for different personnel. For example, for personnel working near the hazard source, due to the higher intensity of the hazard, their cumulative exposure may increase rapidly, and the system will issue an earlier warning time. For personnel far from the hazard source, the system will calculate the time when the hazard and personnel "meet" based on the hazard spread path and the personnel's possible movement trajectory, and issue a timely warning. This warning mechanism based on individual exposure assessment, compared to traditional fixed threshold warning methods, can more accurately determine the time window for hazard occurrence, providing time margin for mine safety management and effectively improving emergency response efficiency.

[0055] In one optional implementation, with the warning time as the constraint objective, the required increase in air volume for the ventilation equipment along the airflow propagation path is calculated in reverse. The wind speed increment corresponding to the air volume value is then substituted into the progressive propagation calculation to obtain the delayed warning time, including: Using the warning time as a time constraint boundary condition, the upper limit of the hazard intensity at each spatial node at the warning time is inferred from the cumulative exposure threshold. The hazard intensity difference that needs to be reduced at each spatial node is calculated by tracing back from the downstream node to the upstream node along the airflow propagation path. Based on the coupling relationship between the difference in hazard intensity and wind speed, a set of air volume optimization equations is established. The node spacing in the airflow propagation topology is used as a constraint parameter. The air volume optimization equations are solved to obtain the air volume increase required for each ventilation device on the airflow propagation path. The wind speed increment of each spatial node is calculated based on the required increase in air volume of the ventilation equipment. The wind speed increment is then superimposed with the original wind speed value in the airflow propagation topology to obtain the adjusted wind speed value. Substitute the adjusted wind speed value into the progressive propagation calculation of diffusion rate and wind speed value to recalculate the adjusted hazard intensity value of each spatial node in the future time sequence; Based on the spatial location of personnel in the future time sequence and the intensity value of the hazard after the control, the cumulative exposure after the control is recalculated, and the moment when the cumulative exposure after the control reaches the cumulative exposure threshold is identified as the delayed warning moment.

[0056] After determining the initial warning time, it is necessary to extend the safe evacuation window through reasonable ventilation control strategies. Specifically, the process of delaying the warning time by adjusting the ventilation system after obtaining the warning time is as follows: Using the warning time as a time constraint boundary condition, and based on the cumulative exposure thresholds for hazardous gases (such as carbon monoxide and sulfur dioxide) or dust concentrations set in mine safety regulations (e.g., cumulative exposure to carbon monoxide not exceeding 350 units within an eight-hour working period), the maximum allowable hazard intensity value for each spatial node at the warning time is derived in reverse, i.e., the upper limit of hazard intensity. In this process, for each node in the airflow propagation topology, starting from the cumulative exposure threshold at the warning time and combining it with the expected dwell time of personnel at each node, the maximum allowable hazard intensity for each node is calculated. Then, following the airflow propagation path, starting from the downstream node (e.g., the end of the return airway), the process traces back upstream to nodes (e.g., the location of the hazard source in the tunneling face or coal mining face). During this backtracking process, based on the difference between the currently predicted hazard intensity value and the calculated upper limit of hazard intensity, the required reduction in hazard intensity at each spatial node is determined.

[0057] An optimization equation set for airflow is established based on the coupling relationship between the difference in hazard intensity and wind speed. In a mine environment, the dilution effect of hazardous substances (such as harmful gases) exhibits a nonlinear relationship with ventilation wind speed, which can generally be expressed as an inverse relationship between hazard intensity and wind speed; that is, the higher the wind speed, the better the dilution effect of the hazardous substance, and the lower the hazard intensity. For each spatial node, a mathematical relationship is established between the wind speed increment and the hazard intensity reduction. Simultaneously, the node spacing in the airflow propagation topology is used as a constraint parameter, as the node spacing determines the time delay of airflow propagation and the degree of hazard concentration attenuation. Furthermore, the physical constraints of the mine ventilation system must be considered, such as the maximum airflow capacity of the ventilation fan and the wind resistance characteristics of the roadway. By solving this multivariate nonlinear equation set, the optimal airflow value that each ventilation device (such as local ventilation fans, auxiliary ventilation ducts, etc.) needs to increase along the airflow propagation path is obtained to achieve the purpose of delaying the warning time.

[0058] Based on the required increase in air volume for the ventilation equipment obtained in the previous step, the wind speed increment at each spatial node is calculated using fluid dynamics principles. In a mine ventilation system, the relationship between air volume and wind speed is influenced by the cross-sectional area of ​​the roadway; wind speed equals air volume divided by the roadway cross-sectional area. For each spatial node, the air volume increment is converted into a corresponding wind speed increment based on the geometric parameters of the roadway (such as cross-sectional area and perimeter) and the characteristics of the ventilation network (such as drag coefficient). Subsequently, these wind speed increments are superimposed with the original wind speed values ​​in the airflow propagation topology to obtain the wind speed values ​​after ventilation control. These wind speed values ​​represent the new wind speed state at each spatial node after the implementation of ventilation control.

[0059] The adjusted wind speed values ​​are substituted into the hazard diffusion model to recalculate the hazard intensity values ​​for each spatial node in the future time series. Based on gas diffusion theory and fluid dynamics principles, the hazard diffusion model can simulate the propagation patterns of hazardous substances under different wind speed conditions. In the progressive propagation calculation, starting from the hazard source, the hazard intensity for each spatial node at each time point is calculated according to the airflow direction. The calculation considers the influence of wind speed on the propagation velocity of hazardous substances, the influence of node spacing on propagation time, and the influence of ventilation conditions on the dilution effect of hazardous substances. Through this calculation, the dynamic changes in hazard intensity for each spatial node in the future time series after adjustment are obtained.

[0060] Based on historical data collected by the personnel positioning system and combined with personnel movement pattern analysis, the spatial distribution of personnel in the future time sequence is predicted. The predicted personnel locations are then combined with the adjusted hazard intensity value calculated in the previous step to calculate the cumulative exposure of each miner during movement. Specifically, the dwell time of each person at each spatial node is multiplied by the adjusted hazard intensity value at that node at the corresponding time, and then accumulated along the time axis to obtain a curve showing the change in cumulative exposure over time. By monitoring when this curve reaches a preset safety threshold, a new warning moment is identified, i.e., the delayed warning moment after adjustment.

[0061] Through the above calculation process, the specific time that the warning time can be delayed after implementing ventilation control measures can be determined. Comparing the delayed warning time with the current real-time time yields a time window suitable for safe evacuation. Based on this time window, and considering the mine roadway network topology, the current location of personnel, and the location of safety exits, the shortest path algorithm is used to plan the optimal evacuation route and generate detailed evacuation instructions. Simultaneously, based on the calculated required increase in airflow for each ventilation device, precise ventilation control instructions are generated to guide timely adjustments to the ventilation system, maximizing the safe evacuation time and ensuring the safety of miners.

[0062] In practical applications, the effectiveness of ventilation control strategies is affected by various factors, such as the response speed of ventilation equipment and the accuracy of control commands. Therefore, the system monitors the actual wind speed changes after ventilation control in real time and compares them with theoretical calculations, making adjustments as necessary to ensure that the ventilation control effect achieves the expected goals, effectively delaying the warning time and buying more time for safe evacuation of personnel.

[0063] Furthermore, in extreme cases, if ventilation control cannot delay the warning time to meet the requirement of safe evacuation of all personnel, the system will automatically activate the emergency plan, such as dispatching additional rescue personnel and evacuating personnel from certain areas in advance, to minimize the potential harm caused by the safety accident. This combination of proactive warning and emergency response significantly improves the intelligence level of mine safety management and emergency response capabilities.

[0064] In one optional implementation, the process of planning evacuation routes based on the time interval between the delayed warning time and the current time, and issuing ventilation control instructions and personnel evacuation instructions, includes: The time difference between the delayed warning time and the current time is calculated as the available evacuation time. Based on the spatial connectivity of the mine roadways, a candidate set of evacuation routes is constructed from the current spatial location of personnel. Based on the personnel movement speed and candidate path length, time-feasible paths with an estimated evacuation time less than the available evacuation time are selected. The hazard intensity value of each spatial node on the time-feasible path after adjustment is obtained. The cumulative exposure of personnel moving along the time-feasible path is calculated. The time-feasible path with the smallest cumulative exposure is selected as the personnel evacuation path. Based on the required increase in air volume for the ventilation equipment, a ventilation control command containing the equipment identification number and the target air volume adjustment value is generated and sent to the control unit. The control unit returns a feedback message indicating that the adjustment is complete. After receiving the feedback message, a personnel evacuation command containing the personnel identification number and the coordinate sequence of the evacuation path is generated based on the personnel evacuation path.

[0065] After determining the delayed warning time, it is necessary to scientifically plan personnel evacuation routes and issue ventilation control instructions and personnel evacuation instructions to relevant systems. Specifically, after obtaining the delayed warning time, the time difference between the delayed warning time and the current system time is calculated. This difference is the available safe evacuation time for personnel. The time difference calculation uses millisecond-level precision to ensure the accuracy of the evacuation plan. For example, if the current system time is T0 and the delayed warning time is T1, then the available evacuation time ΔT = T1 - T0.

[0066] A candidate set of evacuation routes is constructed based on the spatial connectivity graph of mine roadways. The spatial connectivity graph is stored in a graph data structure, where nodes represent key locations such as roadway intersections and work points, and edges represent passable roadway segments. Starting from the personnel's current spatial location, all safety exit locations are set as endpoints. An improved depth-first search algorithm is used to generate multiple candidate evacuation routes from the starting point to each endpoint. To avoid generating too many routes and wasting computational resources, search depth limits and a maximum number of routes are set, typically controlled to within ten. Each route is represented by a sequence of spatial node coordinates.

[0067] The constructed candidate path set is screened for time feasibility. Based on the average movement speed V of personnel and the actual length L of the candidate path, the estimated evacuation time t = L / V is calculated. Movement speeds differ for different types of personnel; a speed parameter database for different personnel types can be established based on historical data. For personnel with mobility impairments, their movement difficulty coefficient must be additionally considered. The estimated evacuation time t is compared with the available evacuation time ΔT, and paths satisfying t < ΔT are selected as the time-feasible path set. If no path meets the condition, the path with the shortest evacuation time is selected, and an emergency support request is issued.

[0068] Obtain the hazard intensity values ​​of each spatial node on the time-feasible path after adjustment. The hazard intensity values ​​are derived from the progressive propagation calculation results in the previous steps. The adjusted values ​​refer to the hazard intensity of each node recalculated after the ventilation equipment increases its airflow. For each node on the path, extract its hazard intensity value at the time when personnel are expected to pass through, forming a hazard intensity sequence for that path.

[0069] The cumulative exposure of personnel moving along a time-feasible path is calculated. For each node on the path, the single-point exposure is calculated by multiplying the expected stay time of the personnel at that node by the hazard intensity value of the node. Then, the single-point exposures of all nodes on the path are summed to obtain the cumulative exposure E. The calculation formula can be expressed as: E=∑(Si×Ti), where Si is the hazard intensity value of node i, and Ti is the stay time of the personnel at node i. In actual calculations, for different types of hazards (such as gas, carbon monoxide, etc.), the degree of hazard of their intensity values ​​is different, and weighting coefficients can be introduced for weighted calculation.

[0070] The path with the lowest cumulative exposure is selected from the set of feasible paths to be the final evacuation route. When the cumulative exposure of multiple paths is similar (the difference is less than a preset threshold), secondary factors such as path length and path complexity can be further considered for a comprehensive decision. Path complexity can be quantitatively assessed through factors such as the number of turns and changes in slope.

[0071] Based on the calculated increase in airflow to the ventilation equipment obtained from previous steps, a ventilation control command is generated. The command includes information such as the equipment identification number, the target airflow value, the control priority, and the control time limit. The equipment identification number is determined by the equipment coding rules of the mine ventilation system; the target airflow value is the increase in airflow based on the current airflow; the control priority is determined based on the extent of the hazard's spread; and the control time limit is set based on the hazard's development rate. The ventilation control command is encapsulated using a standardized data structure and sent to the corresponding control unit via the mine communication network.

[0072] After receiving the ventilation control command, the control unit executes the airflow adjustment operation and returns adjustment status information. This status information includes adjustment progress, actual airflow achieved, and equipment operating parameters. It receives feedback indicating adjustment completion to confirm that the ventilation environment has been adjusted as expected; this is a prerequisite for sending personnel evacuation instructions. If no adjustment completion feedback is received within a preset time, the backup evacuation plan is activated.

[0073] Based on the determined optimal evacuation route, an evacuation instruction is generated. The instruction includes personnel identification numbers, evacuation priorities, evacuation route coordinate sequences, estimated arrival times at each node, and key information along the route. The personnel identification number uniquely identifies the recipient of the evacuation instruction; the evacuation priority is determined based on the degree of hazard at the personnel's location; the evacuation route coordinate sequence is represented as a list of three-dimensional spatial coordinate points; the estimated arrival times at each node are calculated based on the route length and the personnel's movement speed; and the key information includes safety precautions for passing through special areas.

[0074] The generated evacuation instructions are sent to the portable terminal devices of relevant personnel. To ensure the reliability of instruction transmission, a multi-channel redundant transmission mechanism is adopted, including wired communication, wireless communication, and audio-visual signals. After receiving the instructions, the terminal devices display the evacuation route in a graphical interface and guide personnel to evacuate along the designated route through voice prompts. At the same time, the actual movement trajectory of personnel is transmitted back to the central system for real-time monitoring of evacuation progress and dynamic adjustment of the route as necessary.

[0075] During the execution of the instructions, ventilation parameters and personnel location information are continuously monitored. When significant environmental changes or personnel deviating from the planned route are detected, an emergency adjustment mechanism is activated to recalculate the optimal evacuation route and update the evacuation instructions. This dynamic adjustment ensures the adaptability and safety of the evacuation plan in complex and changing environments.

[0076] Taking a mine gas over-limit incident as an example, when the deep learning system detects an abnormal rise in gas concentration in a certain area and a continued tendency to spread, it calculates the warning time as 30 minutes after the current time through airflow propagation. After ventilation control, the warning time is postponed to 45 minutes. The system calculates that the available evacuation time is 45 minutes and generates 5 candidate evacuation routes based on the mine's spatial structure. Considering the average personnel movement speed of 1.2 m / s, 3 time-feasible routes are selected. The cumulative exposure values ​​of these 3 routes are calculated to be 24.5, 18.3, and 29.7 (unit: concentration × minutes), respectively. The route with a cumulative exposure value of 18.3 is selected as the optimal evacuation route. A ventilation control instruction is generated, requiring fans 3, 5, and 7 to increase their airflow by 20%, and sent to the ventilation control system. After receiving feedback that the control is completed, a personnel evacuation instruction is generated and sent to the portable terminals of each worker, guiding them to evacuate safely along the optimal route.

[0077] A second aspect of this invention provides a deep learning-based intelligent early warning and prediction system for mine safety hazards, the system comprising: The first unit is used to acquire monitoring data of the mine production environment, analyze the monitoring data based on a deep neural network, and identify the types and locations of potential hazards. The second unit is used to determine the airflow propagation path from the location of the hidden danger based on the spatial connectivity of the mine roadway and the direction of ventilation airflow, obtain the wind speed value and node spacing of each spatial node on the airflow propagation path, and obtain the airflow propagation topology. The third unit is used to progressively propagate the diffusion rate corresponding to the type of hazard to the wind speed value in the airflow propagation topology, and obtain the hazard intensity value of each spatial node in the future time sequence. The fourth unit is used to predict the spatial location of people in the future time sequence based on their historical movement trajectories, calculate the cumulative exposure based on the time people stay at each location and the hazard intensity value of the corresponding node, and identify the warning time. The fifth unit is used to calculate the required increase in air volume for ventilation equipment along the airflow propagation path with the warning time as the constraint target. The wind speed increment corresponding to the air volume value is substituted into the progressive propagation calculation to obtain the delayed warning time. Based on the time interval between the delayed warning time and the current time, the personnel evacuation path is planned, and ventilation control instructions and personnel evacuation instructions are issued.

[0078] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0079] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0080] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based intelligent early warning and prediction method for mine safety hazards, characterized in that, include: Acquire monitoring data of the mine production environment, analyze the monitoring data based on deep neural networks, and identify the types and locations of potential hazards; Based on the spatial connectivity of mine roadways and the direction of ventilation airflow, the airflow propagation path starting from the location of the hidden danger is determined, and the wind speed value and node spacing of each spatial node on the airflow propagation path are obtained to obtain the airflow propagation topology. The diffusion rate corresponding to the type of hazard is progressively propagated and calculated with the wind speed value in the airflow propagation topology to obtain the hazard intensity value of each spatial node in the future time sequence. Based on the historical movement trajectory of personnel, the spatial location of personnel in the future time sequence is predicted, the cumulative exposure is calculated according to the dwell time of personnel at each location and the hazard intensity value of the corresponding node, and the warning time is identified. Using the warning time as the constraint target, the required increase in air volume for ventilation equipment along the airflow propagation path is calculated in reverse. The wind speed increment corresponding to the air volume value is substituted into the progressive propagation calculation to obtain the delayed warning time. Based on the time interval between the delayed warning time and the current time, personnel evacuation routes are planned, and ventilation control instructions and personnel evacuation instructions are issued.

2. The method according to claim 1, characterized in that, Acquire monitoring data of the mine production environment, analyze the monitoring data based on a deep neural network, and identify the types and locations of potential hazards, including: Collect gas concentration, temperature, and pressure data in the mine production environment, record the data collection time and sensor spatial coordinates, and generate monitoring data with spatiotemporal markers; Abnormal values ​​are removed from the monitoring data, and the data is aligned according to the time of collection. The aligned monitoring data is then normalized by interval transformation to generate standardized monitoring data. Standardized monitoring data is divided into time windows at fixed intervals. Within the time window, convolution operations are performed to extract local features, and nonlinear transformations are applied to the local features to generate feature vectors. A deep neural network is constructed, which includes an input layer, multiple hidden layers, and an output layer. The feature vector is input into the input layer, and through nonlinear transformation and weight transfer in the hidden layer, the probability distribution of the hazard type is calculated in the output layer. The hazard type and hazard severity value are determined based on the probability distribution. The distance matrix is ​​calculated based on the sensor's spatial coordinates. The hazard severity value is converted into a spatial weight. Interpolation calculation is then performed using the distance matrix to determine the spatial location of the hazard.

3. The method according to claim 1, characterized in that, Based on the spatial connectivity and ventilation airflow direction of the mine roadways, the airflow propagation path starting from the location of the potential hazard is determined. The wind speed values ​​and node spacing of each spatial node along the airflow propagation path are obtained, resulting in the airflow propagation topology, including: Collect spatial coordinate points and cross-sectional contour points of the tunnel, calculate the curvature value between the spatial coordinate points, identify the inflection point position based on the curvature value, use piecewise spline interpolation to connect the spatial coordinate points to generate the tunnel center curve, generate the cross-sectional boundary based on the cross-sectional contour points, and combine the tunnel center curve and the cross-sectional boundary to form a spatial connectivity relationship; Collect the power and resistance values ​​of the ventilation equipment, establish an energy transfer equation based on the spatial connectivity, include the surface roughness coefficient and inflection point position in the loss term, solve the energy transfer equation to obtain the velocity field, and determine the ventilation airflow direction based on the velocity field; Starting from the location of the potential hazard, a propagation path map is constructed along the direction of the ventilation airflow. Based on the pressure difference and velocity difference between adjacent spatial nodes in the propagation path map, the propagation probability is calculated to determine the airflow propagation path. The wind speed values ​​of each spatial node on the airflow propagation path are obtained based on the velocity field, the node spacing between spatial nodes is calculated based on the spatial coordinate points, and the flow rate between spatial nodes is calculated in combination with the cross-sectional boundary. The airflow propagation path and the flow rate between spatial nodes are combined to generate an airflow propagation topology.

4. The method according to claim 1, characterized in that, By progressively propagating the diffusion rate corresponding to the hazard type to the wind speed value in the airflow propagation topology, the hazard intensity value of each spatial node in the future time sequence is obtained, including: The diffusion rate of each spatial node at the current moment is calculated based on the diffusion rate corresponding to the type of hazard and the wind speed value in the airflow propagation topology, and a progressive propagation sequence of diffusion amount is constructed. Based on the airflow propagation topology, the connection path and wind direction change path between spatial nodes are determined, the diffusion attenuation and wind speed attenuation between spatial nodes are calculated, the diffusion attenuation and wind speed attenuation are applied to the wind speed value to obtain the corrected wind speed value, and the corrected wind speed value is combined with the diffusion amount progressive propagation sequence to form the hazard propagation parameters. The hazard propagation parameters are transmitted point by point according to the connection sequence of spatial nodes in the airflow propagation topology. The amplitude of the diffusion change at the spatial node is compared with the preset amplitude threshold. The calculation time step is determined based on the comparison result, and the hazard diffusion distribution at the spatial node at each time step is obtained. The spatial node hazard diffusion distribution is unfolded according to the time series. The diffusion change value of adjacent time points and the diffusion accumulation value within the time period are calculated. The initial hazard intensity value is generated by combining them according to the preset ratio. Based on the intensity difference between adjacent time points and the intensity difference between adjacent spatial nodes, outliers are removed and missing values ​​are supplemented to obtain the hazard intensity value of each spatial node in the future time series.

5. The method according to claim 1, characterized in that, Based on historical movement trajectories, the spatial location of individuals in future time sequences is predicted. Cumulative exposure is calculated based on the time individuals spend at each location and the corresponding hazard intensity value at each node. Warning times are identified, including: Collect personnel location tag data to obtain personnel identification number, timestamp and spatial coordinates, and construct a sequence of personnel's historical movement trajectory based on the timestamp order; Calculate the rate of change of acceleration and the rate of change of direction of motion between adjacent timestamps in the historical movement trajectory sequence of personnel, and combine the rate of change of acceleration and the rate of change of direction of motion to form a motion feature sequence; Based on motion feature sequences, the location of dwell points and transition points in the historical movement trajectory sequence of personnel are identified, the dwell time at the dwell point location and the transit time at the transition point location are calculated, and a spatiotemporal distribution sequence is generated. Based on the spatiotemporal distribution sequence, predict the distribution of dwell points and transition points of people in the future time sequence, map the distribution of dwell points and transition points to spatial node locations, and calculate the dwell time at spatial nodes; The exposure intensity is calculated based on the dwell time of the spatial node and the corresponding hazard intensity value. The exposure amount per unit time is calculated based on the exposure intensity according to the time dimension. The cumulative exposure amount is obtained by summing the exposure amounts per unit time. The abrupt change point of the cumulative exposure amount is identified by calculating the rate of change of the cumulative exposure amount as the warning moment.

6. The method according to claim 1, characterized in that, Using the warning time as the constraint objective, the required increase in air volume for ventilation equipment along the airflow propagation path is calculated in reverse. The corresponding wind speed increment is then substituted into the progressive propagation calculation to obtain the delayed warning time, which includes: Using the warning time as a time constraint boundary condition, the upper limit of the hazard intensity at each spatial node at the warning time is inferred from the cumulative exposure threshold. The hazard intensity difference that needs to be reduced at each spatial node is calculated by tracing back from the downstream node to the upstream node along the airflow propagation path. Based on the coupling relationship between the difference in hazard intensity and wind speed, a set of air volume optimization equations is established. The node spacing in the airflow propagation topology is used as a constraint parameter. The air volume optimization equations are solved to obtain the air volume increase required for each ventilation device on the airflow propagation path. The wind speed increment of each spatial node is calculated based on the required increase in air volume of the ventilation equipment. The wind speed increment is then superimposed with the original wind speed value in the airflow propagation topology to obtain the adjusted wind speed value. Substitute the adjusted wind speed value into the progressive propagation calculation of diffusion rate and wind speed value to recalculate the adjusted hazard intensity value of each spatial node in the future time sequence; Based on the spatial location of personnel in the future time sequence and the intensity value of the hazard after the control, the cumulative exposure after the control is recalculated, and the moment when the cumulative exposure after the control reaches the cumulative exposure threshold is identified as the delayed warning moment.

7. The method according to claim 1, characterized in that, Based on the time interval between the delayed warning time and the current time, evacuation routes are planned, and ventilation control instructions and personnel evacuation instructions are issued, including: The time difference between the delayed warning time and the current time is calculated as the available evacuation time. Based on the spatial connectivity of the mine roadways, a candidate set of evacuation routes is constructed from the current spatial location of personnel. Based on the personnel movement speed and candidate path length, time-feasible paths with an estimated evacuation time less than the available evacuation time are selected. The hazard intensity value of each spatial node on the time-feasible path after adjustment is obtained. The cumulative exposure of personnel moving along the time-feasible path is calculated. The time-feasible path with the smallest cumulative exposure is selected as the personnel evacuation path. Based on the required increase in air volume for the ventilation equipment, a ventilation control command containing the equipment identification number and the target air volume adjustment value is generated and sent to the control unit. The control unit returns a feedback message indicating that the adjustment is complete. After receiving the feedback message, a personnel evacuation command containing the personnel identification number and the coordinate sequence of the evacuation path is generated based on the personnel evacuation path.

8. A deep learning-based intelligent early warning and prediction system for mine safety hazards, used to implement the method of any one of claims 1-7, characterized in that, include: The first unit is used to acquire monitoring data of the mine production environment, analyze the monitoring data based on a deep neural network, and identify the types and locations of potential hazards. The second unit is used to determine the airflow propagation path from the location of the hidden danger based on the spatial connectivity of the mine roadway and the direction of ventilation airflow, obtain the wind speed value and node spacing of each spatial node on the airflow propagation path, and obtain the airflow propagation topology. The third unit is used to progressively propagate the diffusion rate corresponding to the type of hazard to the wind speed value in the airflow propagation topology, and obtain the hazard intensity value of each spatial node in the future time sequence. The fourth unit is used to predict the spatial location of people in the future time sequence based on their historical movement trajectories, calculate the cumulative exposure based on the time people stay at each location and the hazard intensity value of the corresponding node, and identify the warning time. The fifth unit is used to calculate the required increase in air volume for ventilation equipment along the airflow propagation path with the warning time as the constraint target. The wind speed increment corresponding to the air volume value is substituted into the progressive propagation calculation to obtain the delayed warning time. Based on the time interval between the delayed warning time and the current time, the personnel evacuation path is planned, and ventilation control instructions and personnel evacuation instructions are issued.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images