Coal mining safety prediction visualization method and system

Abstract

The invention relates to a coal mining safety prediction visualization method and system, and the method comprises the steps: determining a sensor deployment scheme based on mine geological structure information and operation environment characteristics; collecting multi-modal monitoring data according to the sensor deployment scheme, and generating a fusion data set of time-space alignment; extracting corresponding spatio-temporal features, and generating a corresponding risk assessment matrix based on the spatio-temporal features; constructing a four-dimensional tensor field containing a time dimension, and generating a dynamic risk visual image based on the four-dimensional tensor field; in the dynamic risk visual image generation process, receiving a feedback instruction input by an expert, and updating the coal mine safety domain knowledge graph by adopting a fuzzy cognitive mapping mode based on the feedback instruction; and optimizing the space-time convolutional network, updating the dynamic risk visualization image, obtaining an updated risk visualization image result, and executing hierarchical early warning processing. According to the method, the richness and accuracy of coal mining safety prediction visualization can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal mining safety, and in particular relates to a coal mining safety prediction visualization method and system, an electronic device and a non-transitory computer readable storage medium. BACKGROUND

[0002] There are many safety hazards in the process of coal mining, such as gas explosion, coal dust explosion, roof collapse, water inrush, etc. In order to reduce the probability of accidents and improve the safety level of mine operation, the existing technology usually uses sensors to monitor, video inspection, geological modeling and other means to collect key parameters such as gas concentration, temperature and humidity, roof pressure in the mine environment, and combines expert systems, statistical analysis or rule-based models to warn of accident risks. At the same time, some systems will display the monitoring data in the form of charts or heat maps to the dispatch personnel for risk judgment and decision-making assistance.

[0003] However, the data analysis process relies on preset rules or linear models, which is difficult to accurately capture complex risk patterns such as nonlinearity and spatiotemporal coupling. The visualization means are relatively single, often only providing two-dimensional charts or static heat maps, lacking intuitive presentation of the multi-dimensional risk distribution inside the mine, resulting in that the safety management personnel cannot fully grasp the temporal evolution and spatial distribution of potential risks in the decision-making process. SUMMARY

[0004] The present application provides a coal mining safety prediction visualization method, system, electronic device and non-transitory computer readable storage medium that can improve the richness and accuracy of coal mining safety prediction visualization.

[0005] The technical solution of the present application to solve the above technical problems is as follows: The present application provides a coal mining safety prediction visualization method, which comprises: Based on the mine geological structure information and the operation environment characteristics, a sensor deployment scheme is determined through an adaptive layout optimization algorithm; According to the sensor deployment scheme, multi-modal monitoring data is collected, and the multi-modal monitoring data is subjected to heterogeneous data alignment processing driven by quantum information theory to generate a spatiotemporally aligned fusion data set; The spatiotemporally aligned fusion data set is input into a spatiotemporal convolution network with a double-flow attention mechanism to extract corresponding spatiotemporal features, and based on the spatiotemporal features, a corresponding risk assessment matrix is generated in combination with a pre-constructed multi-disaster coupling model; The risk assessment matrix is mapped into a corresponding mine digital twin model to construct a four-dimensional tensor field containing a time dimension, and based on the four-dimensional tensor field, a dynamic risk visualization image is generated through a risk heat map and an isosurface hybrid expression mechanism; In the dynamic risk visualization image generation process, feedback instructions input by experts are received, and a fuzzy cognitive mapping method is used to update the knowledge graph in the coal mine safety field based on the feedback instructions; The spatio-temporal convolution network of the double-flow attention mechanism is optimized using the updated knowledge graph, the dynamic risk visualization image is updated, an updated risk visualization image result is obtained, and hierarchical early warning processing is performed.

[0006] Optionally, the sensor deployment scheme is determined based on the mine geological structure information and the operation environment characteristics through an adaptive layout optimization algorithm, including: A three-dimensional geomechanical grid model is constructed according to the mine geological structure information; Based on the three-dimensional geomechanical grid model, a stress concentration area distribution map of the mine roof is calculated using a finite element stress analysis algorithm; According to the stress concentration area distribution map and the operation environment characteristics, a multi-objective optimization algorithm is used to generate a dynamic layout scheme of the sensor as the sensor deployment scheme.

[0007] Optionally, the multi-modal monitoring data is collected according to the sensor deployment scheme, and a quantum information theory driven heterogeneous data alignment processing is performed on the multi-modal monitoring data to generate a spatio-temporal aligned fusion data set, including: A quantum entangled state correlation matrix is constructed based on the spatio-temporal coordinates of each sensor; Based on the quantum entangled state correlation matrix, a spatio-temporal calibration of gas concentration data and microseismic signals is performed using a quantum mutual information entropy calculation method to obtain calibrated data streams; The calibrated data streams are input into an edge computing node to perform adaptive sampling frequency compression processing to generate the spatio-temporal aligned fusion data set.

[0008] Optionally, the spatio-temporal aligned fusion data set is input into a spatio-temporal convolution network of a double-flow attention mechanism to extract corresponding spatio-temporal features, and based on the spatio-temporal features, a corresponding risk assessment matrix is generated by combining a pre-constructed multi-disaster coupling model, including: The fusion data set is input into the spatio-temporal convolution network to construct a spatial correlation topology graph of the sensor; Based on the spatial correlation topology graph, abnormal fluctuation features in the time dimension and propagation path features in the space dimension are extracted as the spatio-temporal features through a double-flow attention mechanism; The spatio-temporal features are input into a pre-set multi-disaster coupling model, combined with dynamic risk threshold update parameters, and the risk assessment matrix is output.

[0009] Optionally, the risk assessment matrix is mapped into the corresponding mine digital twin model to construct a four-dimensional tensor field containing a time dimension, including: The risk assessment matrix is decomposed into a three-dimensional component containing time, space, and risk intensity; Based on the three-dimensional component, a time-space gradient field of gas diffusion is simulated by a fluid dynamics simulation engine; The time-space gradient field is tensor product operated with the mine digital twin model to construct the four-dimensional risk tensor field.

[0010] Optionally, based on the four-dimensional tensor field, a dynamic risk visualization image is generated by a risk thermodynamic map and isosurface hybrid expression mechanism, including: The four-dimensional tensor field is projected into the coordinate system of the holographic display system; The risk intensity value is mapped into display parameters of two channels of chroma and transparency by a visual perception optimization algorithm; According to the display parameters, the dynamic risk visualization image is generated by a hybrid calculation method of isosurface extraction and thermodynamic map rendering.

[0011] Optionally, during the generation of the dynamic risk visualization image, a feedback instruction input by an expert is received, and based on the feedback instruction, a fuzzy cognitive mapping method is used to update the knowledge graph in the coal mine safety field, including: The abnormal area coordinates contained in the feedback instruction are analyzed and extracted; The abnormal area coordinates are reversely mapped into the risk assessment matrix to generate a corresponding knowledge correction vector; Based on the knowledge correction vector, the disaster coupling weight parameter in the knowledge graph is updated by a fuzzy cognitive mapping mechanism.

[0012] Optionally, the time-space convolutional network of the double-flow attention mechanism is optimized using the updated knowledge graph, the dynamic risk visualization image is updated, and an updated risk visualization image result is obtained, including: According to the updated knowledge graph in the coal mine safety field, a regularization constraint term is constructed; The regularization constraint term is added to the loss function of the multi-scale prediction framework; The network weight parameter is updated by an online incremental learning algorithm, the real-time update of the dynamic risk visualization image is performed, and the updated risk visualization image result is obtained.

[0013] Optionally, the hierarchical early warning processing includes: Based on the updated risk visualization image result, the time-space coordinate information of the corresponding high-risk area is extracted; Based on the space-time coordinates, an optimal evacuation path set in the region is generated by a multi-agent simulation system; The optimal evacuation path set is combined with an emergency resource scheduling model to push corresponding hierarchical warning instructions and corresponding evacuation schemes to the high-risk target region.

[0014] The application also provides a coal mining safety prediction visualization system, the system comprising: A sensor deployment module is configured to determine a sensor deployment scheme based on mine geological structure information and work environment characteristics through an adaptive layout optimization algorithm; A data fusion module is configured to collect multi-modal monitoring data according to the sensor deployment scheme and perform heterogeneous data alignment processing of the multi-modal monitoring data driven by quantum information theory to generate a space-time aligned fusion data set; A risk assessment module is configured to input the space-time aligned fusion data set into a space-time convolution network of a double-flow attention mechanism, extract corresponding space-time features, and generate a corresponding risk assessment matrix based on the space-time features in combination with a pre-constructed multi-disaster coupling model; A visualization module is configured to map the risk assessment matrix to a corresponding mine digital twin model, construct a four-dimensional tensor field containing a time dimension, and generate a dynamic risk visualization image through a risk thermodynamic map and an isosurface hybrid expression mechanism based on the four-dimensional tensor field; A knowledge updating module is configured to receive feedback instructions input by experts during the generation of the dynamic risk visualization image, update a coal mine safety domain knowledge graph in a fuzzy cognitive mapping manner based on the feedback instructions; A safety prediction module is configured to optimize the space-time convolution network of the double-flow attention mechanism using the updated knowledge graph, update the dynamic risk visualization image, obtain an updated risk visualization image result, and perform hierarchical warning processing.

[0015] In addition, to achieve the above-mentioned purpose, the application further provides an electronic device, comprising: a memory for storing a computer software program; a processor for reading and executing the computer software program, thereby realizing a coal mining safety prediction visualization method as described above.

[0016] In addition, to achieve the above-mentioned purpose, the application further provides a non-transitory computer readable storage medium, the storage medium storing a computer software program, the computer software program being executed by a processor to realize a coal mining safety prediction visualization method as described above.

[0017] The application has the following beneficial effects: (1) The present application realizes the reasonable deployment of multiple types of sensors through an adaptive layout optimization algorithm, and can effectively solve the non-synchronous problem of different sensors in time and space dimensions by combining a quantum information theory driven heterogeneous data alignment method, thereby improving the accuracy and consistency of multi-source data fusion.

[0018] (2) The present application adopts a spatio-temporal convolution network with a double-flow attention mechanism, and realizes the collaborative modeling and deep prediction of multiple mine disaster risks (such as gas explosion, roof collapse, etc.) by combining a multi-disaster coupling model and knowledge graph support, thereby significantly improving the identification ability and prediction accuracy of risk evolution trend.

[0019] (3) The present application can dynamically display the risk distribution and change process in the time-space four-dimensional scale by constructing a four-dimensional tensor field in the mine digital twin model and combining a risk thermal map and an isosurface mixed expression mechanism, thereby realizing the stereoscopic and dynamic graphical presentation of the mine safety state and improving the risk situation awareness efficiency.

[0020] In summary, the present application breaks through the technical bottlenecks of single data analysis model, slow response and limited visualization means in the existing coal mine safety monitoring system, and provides a more accurate, intelligent and efficient technical support platform for coal mine safety production, which has significant practical value and popularization prospect. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A flow chart of a coal mining safety prediction visualization method provided by the present application is shown in the figure. Figure 2 A structure schematic diagram of a coal mining safety prediction visualization system provided by the present application is shown in the figure. Figure 3 A hardware structure schematic diagram of a possible electronic device provided by the present application is shown in the figure. Figure 4 A hardware structure schematic diagram of a possible computer readable storage medium provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0023] In the description of the present application, the terms "first", "second" are only for descriptive purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise explicitly specified.

[0024] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or explanation". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given in order to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that those skilled in the art can recognize that the present application can be implemented without using these specific details. In other examples, well-known structures and processes will not be described in detail in order to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope consistent with the principles and features disclosed.

[0025] Please refer to Figure 1 , a flowchart of a coal mining safety prediction visualization method of the present application is provided, including the following steps: Step 201, based on the mine geological structure information and the operation environment characteristics, the sensor deployment scheme is determined by the adaptive layout optimization algorithm.

[0026] In specific implementation, the basic model for deploying sensors can be constructed based on the actual geological structure information and operation environment characteristic parameters of the mine (such as ventilation layout, mining face distribution, etc.). Through the adaptive layout optimization algorithm, the optimal sensor layout scheme is determined under the multi-objective constraints of meeting the monitoring coverage, perception redundancy and layout cost, and the dynamic perception layout planning of the key areas of the mine is realized.

[0027] In some embodiments, step 201 can include: constructing a three-dimensional geomechanical grid model according to the mine geological structure information; based on the three-dimensional geomechanical grid model, using a finite element stress analysis algorithm to calculate the stress concentration area distribution map of the mine roof; According to the stress concentration area distribution map and the operation environment characteristics, a multi-objective optimization algorithm is used to generate a dynamic layout scheme of the sensor as the sensor deployment scheme.

[0028] In the embodiments of the present application, in order to realize accurate perception coverage of the key area of the coal mine, first, a three-dimensional geomechanical grid model is constructed based on the geological structure information of the mine. Specifically, the stratum structure parameters, fault information, rock mass mechanical properties and geological structure data of the area where the mine is located are collected and analyzed, the entire mine space is divided into multiple finite volume units through grid subdivision technology, thereby forming a three-dimensional discretization model with geomechanical properties, which is used to support subsequent mechanical behavior simulation.

[0029] Subsequently, based on the above-mentioned three-dimensional geomechanical grid model, a finite element stress analysis algorithm is used to calculate the stress field of the overburden strata of the mine. In this analysis process, boundary conditions and load models corresponding to the actual working conditions are applied, including gravity loading, surrounding rock stress, mining disturbance and other factors, and the stress tensor distribution in the roof strata is obtained by solving. According to the analysis results, the stress concentration area is identified, that is, the grid unit near the key structure of the mine with high shear stress, high vertical stress or sudden stress gradient, and the stress concentration area distribution map of the mine roof is generated, which reflects the spatial distribution trend of the structure weak area or disaster-prone area.

[0030] On this basis, further combined with the environmental feature information of the mine operation environment, such as the ventilation path, the operation face position, the personnel activity frequency, the existing monitoring point distribution and other environmental constraints, the stress concentration area is taken as the key perception target area, a multi-objective optimization model is constructed, and multiple optimization objectives such as maximizing the monitoring coverage, minimizing the deployment cost and rationalizing the signal redundancy are considered. The layout model is solved by intelligent optimization methods such as particle swarm, genetic algorithm or simulated annealing, and a set of dynamic adaptive sensor deployment scheme is generated as the preposition basis for multi-modal monitoring data acquisition.

[0031] Step 202, according to the sensor deployment scheme, multi-modal monitoring data is collected, and the multi-modal monitoring data is processed by quantum information theory driven heterogeneous data alignment to generate a spatio-temporally aligned fusion data set.

[0032] In a specific implementation, multi-modal monitoring data including gas concentration, temperature and humidity, microseismic, power state, etc. is collected in the deployed sensor network. Due to the heterogeneity of various sensors in sampling frequency, data type and timestamp, a correlation measurement method based on quantum information theory is adopted, and a quantum entangled state correlation matrix and mutual information entropy calculation mechanism are constructed to realize the unified alignment of multi-source data in time and space dimensions, forming a spatio-temporally fused data set with consistent structure and continuous distribution.

[0033] In some embodiments, step 202 can include: constructing a quantum entangled state correlation matrix based on the spatio-temporal coordinates of each sensor; Based on the quantum entanglement state association matrix, the gas concentration data and the microseismic signal are calibrated in time and space by using a quantum mutual information entropy calculation method to obtain a calibrated data stream; The calibrated data stream is input into an edge computing node to perform adaptive sampling frequency compression processing to generate the time-space aligned fusion data set.

[0034] In the embodiment of the present application, in order to realize high-precision time-space fusion processing of heterogeneous multi-modal monitoring data, first, a quantum entanglement state association matrix is constructed based on the time-space coordinate information of various sensor nodes deployed in the mine. The sensors include but are not limited to gas concentration monitoring sensors (such as CH4, CO, CO2 detectors), microseismic event acquisition devices, roof pressure monitoring devices, etc. For each sensor node, its geographic position coordinates in three-dimensional space and its sampling time stamp are extracted to form a multi-dimensional time-space feature vector. Through quantum state modeling method, the feature vector is mapped to the quantum state distribution in the quantum bit system, and further the quantum entanglement state association matrix is constructed by using the superposition and entanglement characteristics of quantum state to depict the cooperative change relationship between different types of sensors in space position and sampling time.

[0035] On the basis of the entanglement state matrix, a quantum mutual information entropy (Quantum Mutual Information Entropy) calculation method is used to perform cross-modal time-space calibration processing on the monitoring data collected from different modalities (especially the gas concentration continuous value and the microseismic signal pulse event). By quantitatively measuring the information coupling degree between the sensor data and combining the mutual dependence coefficients in the entanglement state matrix, the adjustment of the data time stamp and the spatial interpolation reconstruction are realized to solve the asynchronous and mismatch problems caused by inconsistent sampling frequency, communication delay, spatial shielding, etc. The calibrated continuous data stream is output.

[0036] Subsequently, the above calibrated multi-modal data stream is input into the mine edge computing node, and according to the current operation state of the mine, the network load, the sensor working period and other factors, adaptive sampling frequency compression processing is performed. The processing process dynamically adjusts the sampling granularity and data packet merging window of different modal data, reduces the amount of redundant data, improves the time alignment accuracy, and maintains the integrity of the risk feature information. Finally, a time-space aligned fusion data set with a unified time-space reference frame is output, which provides basic data support for subsequent time-space deep learning models.

[0037] Step 203, input the time-space aligned fusion data set into a time-space convolution network of a double-flow attention mechanism, extract corresponding time-space features, and based on the time-space features, generate a corresponding risk assessment matrix in combination with a pre-constructed multi-disaster coupling model.

[0038] In a specific implementation, the fused data set can be input into a dual-stream attention mechanism spatiotemporal convolution network, dynamic fluctuation features are extracted through a time channel, and spatial propagation relationships between sensors are learned through a space channel. On this basis, a pre-constructed coal mine multi-disaster coupling model (for example, a linkage model of gas anomaly and roof damage) is combined to model and analyze potential risk states, and a multi-dimensional risk assessment matrix facing different monitoring nodes and time periods is generated.

[0039] In some embodiments, step 203 can include: inputting the fused data set into the spatiotemporal convolution network to construct a spatial correlation topology graph of the sensors; based on the spatial correlation topology graph, extracting abnormal fluctuation features in the time dimension and propagation path features in the space dimension through a dual-stream attention mechanism as the spatiotemporal features; inputting the spatiotemporal features into a preset multi-disaster coupling model, combining a dynamic risk threshold to update parameters, and outputting the risk assessment matrix.

[0040] In the implementation of the present application, in order to perform deep modeling and prediction analysis on potential risk features in multi-modal fusion monitoring data, the fused data set after spatiotemporal alignment processing is first input into a constructed dual-stream attention mechanism spatiotemporal convolution neural network (Spatiotemporal Convolutional Network with Dual-Stream Attention).

[0041] Specifically, according to the interaction frequency between the geographical position relationship of each sensor node in the fused data set and the perception dimension, a sensor spatial correlation topology graph is constructed. The topology graph takes sensors as graph nodes, takes physical distance, correlation strength, historical event cooperative triggering frequency, etc. as weighted edges, and constitutes a dynamically updated spatial perception graph structure, which is used to reflect the spatial coupling relationship and propagation potential between sensors.

[0042] On the basis of the above spatial topology graph, a dual-stream attention mechanism is used to process feature extraction tasks in the time dimension and the space dimension. Among them, the first stream branch is a time attention channel, which learns dynamic weights for feature sequences at different time steps through a dynamic weight learning mechanism, enhances the response ability to abnormal fluctuations (such as gas concentration mutation and frequent microseisms), and extracts key fluctuation features in the time dimension; the second stream branch is a spatial attention channel, which performs aggregation convolution on the topology graph structure based on a graph convolution kernel operation, captures spatial propagation features such as propagation path, adjacency strength and centrality evolution of disaster information in the sensor network.

[0043] The time and space features extracted in the two branches are then integrated in a multi-dimensional fusion module to form a complete spatio-temporal feature vector as input for disaster prediction and risk reasoning.

[0044] Further, the spatio-temporal feature vector is input into a preset multi-disaster coupling model. The coupling model is a joint reasoning framework constructed based on a data-driven and knowledge graph hybrid modeling method, which can represent the synergistic mechanism between different types of mine disasters (such as gas outburst, roof fall, water inrush, etc.). A dynamic risk threshold updating module is provided inside the model, which can adaptively adjust the risk judgment standard according to the current environmental disturbance degree. Finally, the model outputs a multi-dimensional risk assessment matrix for different regions and time periods, where each matrix element reflects the disaster occurrence probability or severity score at a specific spatio-temporal location.

[0045] The risk assessment matrix serves as the core basis for subsequent visualization modeling and early warning control of the application, enabling accurate identification and dynamic prediction of multi-disaster interaction coupling risks in complex underground environments.

[0046] In step 204, the risk assessment matrix is mapped into the corresponding mine digital twin model to construct a four-dimensional tensor field containing the time dimension. Based on the four-dimensional tensor field, dynamic risk visualization images are generated through a risk heat map and isosurface hybrid expression mechanism.

[0047] In specific implementations, the risk assessment matrix can be mapped and embedded into a mine three-dimensional digital twin model, with the addition of a time dimension to construct a four-dimensional tensor field containing space, risk intensity, and time evolution. This tensor field is expressed through a risk heat map rendering and isosurface extraction algorithm, allowing dynamic risk information to be visually presented in various visualization dimensions such as volume rendering, transparent channel, and color gradient, enhancing the perceptibility of mine operating conditions.

[0048] In some embodiments, step 204 can include: decomposing the risk assessment matrix into three-dimensional components containing time, space, and risk intensity; simulating the spatio-temporal gradient field of gas diffusion through a fluid dynamics simulation engine based on the three-dimensional components; performing tensor product operations on the spatio-temporal gradient field and the mine digital twin model to construct the four-dimensional risk tensor field.

[0049] In the implementation of the application, to further realize dynamic modeling and high-dimensional information expression of the mine risk evolution process, a four-dimensional risk tensor field with time continuity and spatial distribution needs to be constructed based on the previously generated risk assessment matrix.

[0050] Specifically, the risk assessment matrix is first processed by tensor decomposition in time dimension, space dimension, and risk intensity dimension to obtain a three-dimensional component set with a clear structure. The time dimension component reflects the evolution trend of the risk index in the continuous time sequence; the space dimension component represents the geographical distribution characteristics between different sensor nodes or mine sub-regions; and the risk intensity component records the disaster level, occurrence probability, or other quantitative risk index values at the corresponding position and time node. The three-dimensional structure constitutes a preliminary multi-dimensional data basis.

[0051] Based on the above three-dimensional components, a fluid dynamics simulation engine is further introduced to more realistically simulate the diffusion characteristics of harmful gases (such as CH4 and CO) in complex underground tunnel systems. The simulation engine is based on the finite volume method or the lattice Boltzmann method (LBM), combined with boundary conditions such as mine gas flow structure parameters, gas emission velocity, ventilation rate, and temperature and pressure conditions, to numerically simulate the diffusion rate and direction of gas in geographical space, and further construct a spatio-temporal gradient field of gas diffusion. The gradient field can represent the concentration change rate, transmission path, and time delay effect of the risk substance.

[0052] Subsequently, the spatio-temporal gradient field and the digital twin model of the mine are subjected to tensor product operation. The digital twin model is a three-dimensional simulation mapping of the actual physical space of the mine, containing various engineering information such as structure geometry, geological stratification, operation arrangement, and ventilation system. By performing tensor product operation on the data tensor of the spatio-temporal gradient field and the structure tensor of the digital twin model in the common space dimension, spatial mapping and dynamic embedding of risk intensity are realized, thereby constructing a four-dimensional risk tensor field that integrates time, space, physical structure, and disaster dynamics.

[0053] The tensor field serves as a key input for subsequent dynamic visualization rendering and intelligent early warning decision-making, providing information foundation and data support for real-time perception and multi-dimensional linkage of coal mine safety situation.

[0054] In some embodiments, step 204 can further include: projecting the four-dimensional tensor field to the coordinate system of the holographic display system; mapping the risk intensity values to display parameters of two channels of chroma and transparency through a visual perception optimization algorithm; generating the dynamic risk visualization image through a mixed calculation method of isosurface extraction and heat map rendering according to the display parameters.

[0055] In the implementation of the present application, in order to realize the multi-dimensional expression and dynamic visualization interaction of the risk situation of multiple disaster types in the coal mine underground, a visual expression image for human-computer understanding needs to be generated based on the aforementioned four-dimensional risk tensor field. To this end, first, the four-dimensional tensor field is mapped to the three-dimensional coordinate system of the holographic display system, completing the projection conversion from the data space to the visual space.

[0056] Specifically, the four-dimensional risk tensor field contains four dimensions of time, spatial position (x, y, z), and risk intensity. In order to realize dynamic display in a three-dimensional visual environment, the tensor field is continuously sliced along the time axis, and the three-dimensional tensor data at each time point is projected into the visual coordinate system of the holographic display system in turn. The display system supports a virtual mine model constructed based on a real spatial scale, has the ability of three-dimensional panoramic view angle control and depth perception enhancement, and can realize stereoscopic dynamic visual expression.

[0057] In order to enhance the readability and intuitive perception effect of the risk image, the risk intensity values in the tensor field are further mapped and processed based on a visual perception optimization algorithm. Specifically, a perception-driven mapping function is used to normalize the risk intensity values to a preset interval, and the normalized values are mapped to two display channel parameters, namely the hue channel and the alpha channel. The hue channel is used to indicate the category change of different risk levels, and the alpha channel is used to express the significance of the risk intensity, thereby realizing the joint control of risk level and visual penetration.

[0058] After the above mapping is completed, a hybrid calculation method based on isosurface extraction and heat map rendering is further used to generate dynamic risk visualization images. Among them: The isosurface extraction uses the Marching Cubes or Marching Tetrahedra algorithm to construct a three-dimensional isosurface for the spatial region whose risk intensity value is in a specific isosurface interval, forming a risk entity with a shape structure; The heat map rendering uses pseudo-color coding and multi-layer transparent superposition technology to perform color transition rendering according to the risk gradient in the external area of the isosurface, highlighting the risk propagation trend and concentration change characteristics.

[0059] Finally, the dynamic risk visualization image will be displayed in the form of a three-dimensional dynamic layer in the holographic display system, with interactive functions such as rotation, scaling, and time sequence backtracking, providing high-dimensional, intuitive, and continuous risk cognition support for dispatch personnel and expert users.

[0060] Step 205, in the process of generating the dynamic risk visualization image, receiving feedback instructions input by experts, and updating the knowledge graph in the field of coal mine safety based on the feedback instructions using a fuzzy cognitive mapping method.

[0061] In a specific implementation, during the dynamic visualization process, the system supports human experts to issue intervention instructions for abnormal areas or suspicious situations, and the feedback results will be parsed as knowledge correction information. Through the fuzzy cognitive mapping method, the feedback data is converted into the update logic of the weight between nodes in the graph, realizing the continuous optimization of the disaster linkage relationship in the coal mine safety knowledge graph.

[0062] In some embodiments, step 205 can include: parsing and extracting the abnormal area coordinates contained in the feedback instructions; mapping the abnormal area coordinates to the risk assessment matrix in reverse to generate a corresponding knowledge correction vector; Based on the knowledge correction vector, update the disaster coupling weight parameters in the knowledge graph through the fuzzy cognitive mapping mechanism.

[0063] In the implementation of the present application, in order to realize the dynamic absorption of human expert knowledge by the coal mine risk prediction system and the model self-adaptive update, an expert interactive feedback mechanism needs to be introduced in the process of generating dynamic risk visualization images, and the knowledge graph in the field of coal mine safety is optimized in real time based on this mechanism.

[0064] Specifically, when the dynamic risk visualization image is presented in the holographic display system, the system receives feedback instructions from expert users, which include information such as high-risk areas, false positive areas, or potential hazard areas labeled by experts based on risk perception experience. The system performs semantic analysis and structured processing on the feedback instructions to extract the abnormal area coordinate information contained therein. The abnormal area coordinates can represent their spatial position in the mine model in the form of three-dimensional space coordinates (x, y, z), and can be further bound to corresponding timestamp, disaster type label and other meta information.

[0065] Subsequently, the extracted abnormal area coordinates are mapped to the risk assessment matrix generated by the system through a spatial reflection mapping mechanism. This reverse mapping operation is used to locate the index position of the abnormal area in the original tensor field and extract the corresponding risk feature response value. Based on the extraction result, a knowledge correction vector consistent with the feedback content is constructed. The knowledge correction vector represents the deviation area and direction between expert judgment and model evaluation, and is a key data carrier for guiding the adjustment of the knowledge graph parameters.

[0066] After the knowledge correction vector is generated, a fuzzy cognitive mapping mechanism is introduced to dynamically update the disaster coupling relationships involved in the knowledge graph. Specifically, the system uses a cognitive mapping algorithm based on a fuzzy relationship matrix, takes the knowledge correction vector as an input signal, and updates the causal relationship strength values, mutual influence weights, and synergistic trigger probabilities between disasters in the knowledge graph through steps such as fuzzy weight propagation, error-driven adjustment, and multi-dimensional fusion mapping, to form a knowledge-enhanced representation oriented to the current mine environment.

[0067] In the above manner, semantic conversion of expert feedback, data reverse mapping, and structure optimization of the knowledge layer are realized, enabling the present application to have continuous learning and self-adaptive evolution capabilities, providing a knowledge support foundation for subsequent prediction model optimization and risk image updating.

[0068] Step 206, using the updated knowledge graph to optimize the spatio-temporal convolution network of the double-flow attention mechanism, updating the dynamic risk visualization image, obtaining the updated risk visualization image result, and performing hierarchical warning processing.

[0069] In specific implementations, a regularization term can be constructed based on the updated knowledge graph to constrain the training loss function of the spatio-temporal convolution network, and the network weights are updated through online incremental learning, thereby improving the model's adaptability to dynamic environments. Subsequently, the system identifies high-risk areas and determines their levels based on the latest risk visualization image results, triggering targeted hierarchical warning and emergency response measures.

[0070] In some embodiments, step 206 can include: constructing a regularization constraint term based on the updated knowledge graph of the coal mine safety field; adding the regularization constraint term to the loss function of the multi-scale prediction framework; updating the network weight parameters using an online incremental learning algorithm to update the dynamic risk visualization image in real time, obtaining the updated risk visualization image result.

[0071] In the implementation of the present application, to effectively absorb the feedback of the prediction model to the knowledge graph and further improve the accuracy and real-time performance of risk prediction and visualization, a knowledge-driven network structure optimization mechanism is proposed. This mechanism combines regularization constraint construction and online incremental learning to dynamically adjust the weights of the spatio-temporal convolution network of the double-flow attention mechanism, thereby driving the real-time updating of the dynamic risk visualization image.

[0072] Specifically, first, according to the updated knowledge graph in the field of coal mine safety, knowledge elements reflecting multi-disaster coupling relationship, spatial propagation mode, time evolution characteristics and the like are extracted, and the knowledge elements are converted into network analyzable data representation through a graph structure embedding and attribute vectorization method. Based on the representation result, a regularization constraint term is constructed to guide the network learning result to be close to the prior structure contained in the knowledge graph during model training.

[0073] The regularization constraint term can take the form of a graph structure conforming loss function, a relationship maintaining constraint function or a disaster coupling weight guiding function, and the core goal is to maintain the structural consistency and semantic rationality of the neural network in the prediction output space. Specifically, the constraint term is embedded into the loss function of the multi-scale prediction framework, and works together with the original prediction error loss (such as mean square error, cross entropy loss) to form a total loss function for model updating.

[0074] Then, the system uses an online incremental learning algorithm to dynamically update the weight parameters of the spatio-temporal convolution network. The algorithm, on the basis of maintaining the stability of the learned parameters, fuses new knowledge samples and structural constraint information, optimizes the network level weight and attention parameter in real time, and ensures that the model has the ability to continuously adapt to the changes in the mine environment. Through periodic or event-driven mode, the network updating process is triggered, and the continuous refreshing of the dynamic risk visualization image is driven.

[0075] Finally, based on the latest risk assessment results output by the optimized model, the four-dimensional risk tensor field is reconstructed and the dynamic visualization image is updated, forming an updated risk visualization image result highly consistent with the current mine state and expert knowledge, providing more timely and accurate visual cognitive support for emergency decision-making.

[0076] In some embodiments, step 206 can further include: extracting the spatio-temporal coordinate information of the corresponding high-risk area based on the updated risk visualization image result; generating an optimal evacuation path set in the area through a multi-agent simulation system based on the spatio-temporal coordinates; pushing corresponding hierarchical warning instructions and corresponding evacuation schemes to the high-risk target area by combining the optimal evacuation path set with an emergency resource scheduling model.

[0077] In the implementation process of the present application, in order to realize intelligent response and efficient disposal under the condition of mine sudden risk, the system constructs a hierarchical warning and evacuation decision mechanism for the target area based on the dynamic updated risk visualization image result. The mechanism integrates multi-agent simulation and resource scheduling optimization strategies, realizes rapid identification of high-risk areas, evacuation path planning and warning instruction generation, and significantly improves the timeliness and intelligence level of coal mine disaster response.

[0078] Specifically, the system first analyzes the updated risk visualization image results, extracts the regions with risk levels exceeding the preset threshold using an image segmentation algorithm and a tensor threshold determination method, and obtains the spatiotemporal coordinate information of the high-risk regions in the mine three-dimensional model in combination with the holographic coordinate mapping relationship. The coordinate information includes a spatial position (three-dimensional coordinates) and a corresponding time label, which are used to describe the spatial range and evolution trend of the risk region.

[0079] Then, based on the extracted spatiotemporal coordinate information, a preconfigured multi-agent simulation system is called. The simulation system simulates multi-path escape behavior in the high-risk region by constructing a mine passage topology graph and a risk avoidance node network, considers factors such as personnel density, ventilation conditions, obstacle distribution, and disaster diffusion speed, and comprehensively evaluates the path length, safety factor, and evacuation efficiency of multiple feasible evacuation paths. Finally, a set of optimal evacuation paths is output, and the path set includes multiple dynamic path schemes that are verified to be reachable, safe, and risk-avoiding.

[0080] After obtaining the set of optimal evacuation paths, the system further combines an emergency resource scheduling model of the coal mine site. The scheduling model includes information such as the location of existing emergency supplies in the mine, the state of rescue passages, and the deployable rescue forces, and through an optimization allocation algorithm, a hierarchical response matching strategy between the risk region and the emergency response resources is constructed.

[0081] Based on the above path set and scheduling model, the system generates corresponding hierarchical warning instructions, including but not limited to alarm level division, emergency broadcast content, control center instructions, and individualized evacuation schemes for different work areas. The warning instructions and evacuation schemes can be pushed to target area workers and management personnel through wearable terminals, command center large screens, mine broadcast systems, and other forms, ensuring the coverage and operability of the warning response.

[0082] Through the above technical process, the present application not only realizes accurate visualization of disaster situation, but also completes an intelligent closed loop from risk perception to warning action, providing a real-time, scientific, and efficient decision support means for coal mining safety.

[0083] Please refer to Figure 2 , Figure 2 The structure diagram of a coal mining safety prediction visualization system provided by the present application is shown.

[0084] As Figure 2 shown, the coal mining safety prediction visualization system provided by the present application includes: A sensor deployment module 301 is used to determine a sensor deployment scheme based on mine geological structure information and work environment characteristics through an adaptive layout optimization algorithm. The data fusion module 302 is used to collect multimodal monitoring data according to the sensor deployment scheme, and perform quantum information theory-driven heterogeneous data alignment processing on the multimodal monitoring data to generate a spatiotemporally aligned fusion dataset. The risk assessment module 303 is used to input the spatiotemporally aligned fusion dataset into a spatiotemporal convolutional network with a dual-stream attention mechanism, extract the corresponding spatiotemporal features, and generate the corresponding risk assessment matrix based on the spatiotemporal features and a pre-built multi-hazard coupling model. The visualization module 304 is used to map the risk assessment matrix to the corresponding mine digital twin model, construct a four-dimensional tensor field containing the time dimension, and generate dynamic risk visualization images based on the four-dimensional tensor field through a risk heat map and isosurface hybrid expression mechanism. The knowledge update module 305 is used to receive feedback instructions input by experts during the generation of the dynamic risk visualization image, and update the knowledge graph in the field of coal mine safety based on the feedback instructions using a fuzzy cognitive mapping method. The safety prediction module 306 is used to optimize the spatiotemporal convolutional network of the dual-stream attention mechanism using the updated knowledge graph, update the dynamic risk visualization image, obtain the updated risk visualization image result, and perform hierarchical early warning processing.

[0085] Please see Figure 3 , Figure 3 This is a schematic diagram illustrating an embodiment of the electronic device provided in this invention. For example... Figure 3 As shown, an embodiment of the present invention provides an electronic device 400, including a memory 410, a processor 420, and a computer program 411 stored in the memory 410 and executable on the processor 420. When the processor 420 executes the computer program 411, it performs the following steps: Based on mine geological structure information and operating environment characteristics, an adaptive layout optimization algorithm is used to determine the sensor deployment scheme. Multimodal monitoring data is collected according to the sensor deployment scheme, and the multimodal monitoring data is subjected to quantum information theory-driven heterogeneous data alignment processing to generate a spatiotemporally aligned fusion dataset. The spatiotemporally aligned fusion dataset is input into a spatiotemporal convolutional network with a dual-stream attention mechanism to extract corresponding spatiotemporal features. Based on the spatiotemporal features, combined with a pre-built multi-hazard coupling model, a corresponding risk assessment matrix is ​​generated. The risk assessment matrix is ​​mapped to the corresponding mine digital twin model to construct a four-dimensional tensor field containing the time dimension. Based on the four-dimensional tensor field, a dynamic risk visualization image is generated through a risk heat map and isosurface hybrid expression mechanism. In the dynamic risk visualization image generation process, feedback instructions input by experts are received, and a fuzzy cognitive mapping method is used to update the knowledge graph in the coal mine safety field based on the feedback instructions; The spatio-temporal convolution network of the double-flow attention mechanism is optimized using the updated knowledge graph, the dynamic risk visualization image is updated, an updated risk visualization image result is obtained, and hierarchical early warning processing is performed.

[0086] Please refer to Figure 4 , Figure 4 An embodiment of a computer readable storage medium provided by the embodiment of the present application is shown in the figure. Figure 4 As shown in the figure, the embodiment provides a computer readable storage medium 500, which stores a computer program 411, and the computer program 411 is executed by a processor to implement the following steps: Based on the mine geological structure information and the operation environment characteristics, a sensor deployment scheme is determined through a self-adaptive layout optimization algorithm; According to the sensor deployment scheme, multi-modal monitoring data is collected, and a quantum information theory driven heterogeneous data alignment processing is performed on the multi-modal monitoring data to generate a spatio-temporal aligned fusion data set; The spatio-temporal aligned fusion data set is input into a spatio-temporal convolution network of a double-flow attention mechanism, corresponding spatio-temporal features are extracted, and based on the spatio-temporal features, a corresponding risk assessment matrix is generated by combining a pre-constructed multi-disaster coupling model; The risk assessment matrix is mapped to a corresponding mine digital twin model, a four-dimensional tensor field containing a time dimension is constructed, and based on the four-dimensional tensor field, a dynamic risk visualization image is generated through a risk thermodynamic map and an isosurface hybrid expression mechanism. In the dynamic risk visualization image generation process, feedback instructions input by experts are received, and a fuzzy cognitive mapping method is used to update the knowledge graph in the coal mine safety field based on the feedback instructions; The spatio-temporal convolution network of the double-flow attention mechanism is optimized using the updated knowledge graph, the dynamic risk visualization image is updated, an updated risk visualization image result is obtained, and hierarchical early warning processing is performed.

[0087] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0088] Those skilled in the art will appreciate that embodiments of the present application can be devised for a variety of applications. It is intended that the present application be limited only by the scope of the appended claims, and it is intended that various modifications and alterations made by those skilled in the art be considered as within the scope of the present application. The embodiments of the present application will be described with reference to the attached drawings identified below.

[0089] The present application is described in reference to the drawings using a flowchart and / or a block diagram of the method, apparatus (system) and computer program product according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0090] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0091] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0092] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments by those skilled in the art once they learn of the basic inventive concepts. Therefore, the appended claims are intended to cover all such modifications and variations as fall within the scope of the present application.

[0093] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A method for visualizing and predicting safety in coal mining, characterized in that, The method includes: Based on mine geological structure information and operating environment characteristics, an adaptive layout optimization algorithm is used to determine the sensor deployment scheme. Multimodal monitoring data is collected according to the sensor deployment scheme, and the multimodal monitoring data is subjected to quantum information theory-driven heterogeneous data alignment processing to generate a spatiotemporally aligned fusion dataset. The spatiotemporally aligned fusion dataset is input into a spatiotemporal convolutional network with a dual-stream attention mechanism to extract corresponding spatiotemporal features. Based on the spatiotemporal features, combined with a pre-built multi-hazard coupling model, a corresponding risk assessment matrix is ​​generated. The risk assessment matrix is ​​mapped to the corresponding mine digital twin model to construct a four-dimensional tensor field containing the time dimension. Based on the four-dimensional tensor field, a dynamic risk visualization image is generated through a risk heat map and isosurface hybrid expression mechanism. During the generation of the dynamic risk visualization image, feedback instructions from experts are received, and the knowledge graph in the field of coal mine safety is updated based on the feedback instructions using a fuzzy cognitive mapping method. The spatiotemporal convolutional network of the dual-stream attention mechanism is optimized using the updated knowledge graph to update the dynamic risk visualization image, resulting in an updated risk visualization image, and then a graded early warning process is performed.

2. The method according to claim 1, characterized in that, The process of determining sensor deployment schemes based on mine geological structure information and operational environment characteristics using an adaptive layout optimization algorithm includes: A three-dimensional geomechanical mesh model is constructed based on the geological structure information of the mine. Based on the aforementioned three-dimensional geomechanical grid model, the stress concentration area distribution map of the mine roof is calculated using the finite element stress analysis algorithm; Based on the stress concentration area distribution map and the characteristics of the working environment, a dynamic layout scheme for the sensors is generated using a multi-objective optimization algorithm as the sensor deployment scheme.

3. The method according to claim 2, characterized in that, The process of collecting multimodal monitoring data according to the sensor deployment scheme and performing quantum information theory-driven heterogeneous data alignment processing on the multimodal monitoring data to generate a spatiotemporally aligned fusion dataset includes: A quantum entangled state correlation matrix is ​​constructed based on the spatiotemporal coordinates of each sensor; Based on the quantum entangled state correlation matrix, the gas concentration data and microseismic signals are spatiotemporally calibrated using the quantum mutual information entropy calculation method to obtain the calibrated data stream. The calibrated data stream is input to the edge computing node, where adaptive sampling frequency compression is performed to generate the spatiotemporally aligned fusion dataset.

4. The method according to claim 3, characterized in that, The process involves inputting the spatiotemporally aligned fused dataset into a spatiotemporal convolutional network with a two-stream attention mechanism to extract corresponding spatiotemporal features. Based on these features, and combined with a pre-built multi-hazard coupling model, a corresponding risk assessment matrix is ​​generated, including: The fused dataset is input into the spatiotemporal convolutional network to construct a spatial association topology map of the sensors; Based on the aforementioned spatial correlation topology graph, the abnormal fluctuation features in the time dimension and the propagation path features in the spatial dimension are extracted as the spatiotemporal features through a two-stream attention mechanism. The spatiotemporal features are input into a preset multi-hazard coupling model, and combined with dynamic risk threshold update parameters, the risk assessment matrix is ​​output.

5. The method according to claim 4, characterized in that, The process of mapping the risk assessment matrix to the corresponding mine digital twin model and constructing a four-dimensional tensor field including the time dimension includes: The risk assessment matrix is ​​decomposed into three-dimensional components that include time, space, and risk intensity. Based on the aforementioned three-dimensional components, the spatiotemporal gradient field of gas diffusion is simulated using a fluid dynamics simulation engine; The spatiotemporal gradient field is combined with the digital twin model of the mine using tensor product operation to construct the four-dimensional risk tensor field.

6. The method according to claim 5, characterized in that, The generation of dynamic risk visualization images based on the four-dimensional tensor field, through a hybrid expression mechanism of risk heatmap and isosurface, includes: The four-dimensional tensor field is projected onto the coordinate system of the holographic display system; Through a visual perception optimization algorithm, the risk intensity value is mapped to display parameters for the chroma and transparency channels, respectively. Based on the displayed parameters, the dynamic risk visualization image is generated through a hybrid calculation method combining isosurface extraction and heatmap rendering.

7. The method according to claim 6, characterized in that, During the generation of the dynamic risk visualization image, feedback instructions are received from experts. Based on these instructions, the knowledge graph in the field of coal mine safety is updated using a fuzzy cognitive mapping method, including: Parse and extract the coordinates of the abnormal area contained in the feedback instruction; The coordinates of the abnormal region are mapped inversely to the risk assessment matrix to generate a corresponding knowledge correction vector; Based on the knowledge correction vector, the disaster coupling weight parameters in the knowledge graph are updated through a fuzzy cognitive mapping mechanism.

8. The method according to claim 7, characterized in that, The process of optimizing the spatiotemporal convolutional network of the dual-stream attention mechanism using the updated knowledge graph to update the dynamic risk visualization image and obtain the updated risk visualization image result includes: Construct regularization constraints based on the updated knowledge graph of coal mine safety. The regularization constraint term is added to the loss function of the multi-scale prediction framework; An online incremental learning algorithm is used to update the network weight parameters, thereby updating the dynamic risk visualization image in real time and obtaining the updated risk visualization image result.

9. The method according to claim 8, characterized in that, The execution of tiered early warning processing includes: Based on the updated risk visualization image results, extract the spatiotemporal coordinate information of the corresponding high-risk areas; Based on the aforementioned spatiotemporal coordinates, a set of optimal evacuation paths within the region is generated using a multi-agent simulation system. By combining the set of optimal evacuation routes with the emergency resource scheduling model, corresponding graded early warning instructions and corresponding evacuation plans are pushed to the high-risk target areas.

10. A coal mine safety prediction and visualization system, characterized in that, The system includes: The sensor deployment module is used to determine the sensor deployment scheme based on mine geological structure information and working environment characteristics through an adaptive layout optimization algorithm. The data fusion module is used to collect multimodal monitoring data according to the sensor deployment scheme, and to perform quantum information theory-driven heterogeneous data alignment processing on the multimodal monitoring data to generate a spatiotemporally aligned fusion dataset. The risk assessment module is used to input the spatiotemporally aligned fusion dataset into a spatiotemporal convolutional network with a dual-stream attention mechanism, extract the corresponding spatiotemporal features, and generate the corresponding risk assessment matrix based on the spatiotemporal features and a pre-built multi-hazard coupling model. The visualization module is used to map the risk assessment matrix to the corresponding mine digital twin model, construct a four-dimensional tensor field containing the time dimension, and generate dynamic risk visualization images based on the four-dimensional tensor field through a risk heat map and isosurface hybrid expression mechanism. The knowledge update module is used to receive feedback instructions from experts during the generation of the dynamic risk visualization image, and update the knowledge graph in the field of coal mine safety based on the feedback instructions using a fuzzy cognitive mapping method. The safety prediction module is used to optimize the spatiotemporal convolutional network of the dual-stream attention mechanism using the updated knowledge graph, update the dynamic risk visualization image, obtain the updated risk visualization image result, and perform hierarchical early warning processing.

Images