An intelligent safety prevention and control system for multiple disasters in a coal mine underground

Abstract

The application discloses a coal mine underground multi-disaster intelligent safety prevention and control system, and belongs to the technical field of coal mine safety monitoring and intelligent management and control. The system comprises a multi-dimensional dynamic twin body construction module, which is used for constructing a three-dimensional static geological model according to geological exploration data, and dynamically updating the three-dimensional static geological model according to underground internet of things real-time monitoring data and mining progress data, so as to form a 4D dynamic twin body which evolves in real time with the advancing of a mining working face; and a multi-disaster coupling decoupling analysis module based on a graph neural network, which is used for taking a mining space grid unit as a node, taking monitoring data of multiple disasters as node attributes, and taking a spatial proximity relationship and a physical action relationship between nodes as edges to construct a disaster coupling relationship graph model; the disaster coupling relationship graph model is subjected to inference calculation through a pre-trained graph neural network; and the multi-disaster coupling decoupling analysis, dominant disaster element identification and virtual governance deduction integration purposes can be achieved.

Description

Technical Field

[0001] This invention belongs to the field of coal mine safety monitoring and intelligent management technology, specifically relating to an intelligent safety prevention and control system for multiple disasters in underground coal mines. Background Technology

[0002] In recent years, with the continuous advancement of digital mine construction, related technologies have made significant progress. Technologies such as 3D geological modeling, IoT sensing and monitoring, and big data analysis have been gradually applied in the field of coal mine safety, providing strong support for the early identification and warning of disasters.

[0003] However, existing technological systems still have significant shortcomings in addressing the coupling effects of multiple hazards. Specifically, these shortcomings include: weak multi-source data fusion and analysis capabilities, with massive amounts of monitoring data failing to be effectively integrated and deeply mined, leading to unclear identification of coupling relationships between hazards; 3D geological models are mostly built based on static data, lacking dynamic correlation and updates with real-time production data, making it difficult to reflect changes in underground geological conditions in a timely manner; a lack of systematic quantitative analysis methods for the coupling mechanisms of multiple hazards such as gas, rockburst, and water hazards, making it difficult to identify the dominant hazard elements and their evolution paths; furthermore, the formulation of disaster management plans still relies heavily on experience-based judgments, lacking means to conduct pre-simulation, comparison, and effect evaluation in the digital space, making it difficult to achieve proactive and precise risk prevention and control.

[0004] Therefore, in order to address the aforementioned technical issues, it is necessary to provide an intelligent safety control system for multiple disasters in underground coal mines. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent safety control system for multiple disasters in coal mines that integrates multi-hazard coupling and decoupling analysis, dominant disaster element identification, and virtual governance simulation.

[0006] To achieve the above objectives, the present invention provides an intelligent safety control system for multiple hazards in underground coal mines, comprising:

[0007] The multidimensional dynamic twin construction module is used to construct a three-dimensional static geological model based on geological exploration data, and dynamically update the three-dimensional static geological model based on real-time monitoring data from the underground Internet of Things and mining progress data, forming a 4D dynamic twin that evolves in real time as the mining face advances;

[0008] The multi-hazard coupling and decoupling analysis module based on graph neural networks is used to construct a disaster coupling relationship graph model with mining space grid cells as nodes, monitoring data of multiple disasters as node attributes, and spatial proximity and physical interaction relationships between nodes as edges. The disaster coupling relationship graph model is inferred and calculated by a pre-trained graph neural network to quantify the coupling strength coefficient and disaster contribution of each disaster element, and to identify the dominant disaster element and its associated secondary disaster chain in the current mining environment based on the calculation results.

[0009] The virtual space decoupling simulation module is used to generate at least one virtual decoupling governance scheme for the identified dominant disaster element, and to simulate the disaster evolution process after implementing the virtual decoupling governance scheme in the 4D dynamic twin, and output chain break suggestions to guide on-site governance.

[0010] In one or more embodiments of the present invention, the multidimensional dynamic twin construction module includes:

[0011] The static geological modeling unit is used to construct a three-dimensional static geological model containing information on coal seams, faults, collapse columns, and aquifers based on borehole data, seismic exploration data, and well logging data.

[0012] The real-time data assimilation unit is used to access the real-time monitoring data of the Internet of Things and use a discrete smooth interpolation algorithm or a point-layer thickness grid model that integrates vector and grid to assimilate the mining exposure information and monitoring data into the three-dimensional static geological model.

[0013] The dynamic update engine is used to trigger local updates of the model based on mining progress data, so as to realize the geological model and production progress in line with mining.

[0014] In one or more embodiments of the present invention, the multi-hazard coupling decoupling analysis module based on graph neural networks includes:

[0015] The disaster coupling relationship graph construction unit is used to divide the mining space into multiple grid units as nodes of the graph model, and to use one or more monitoring data of gas concentration, ground stress value, water pressure, and temperature as node attributes. At the same time, the connection edges between nodes are established according to spatial distance and physical and mechanical properties of rock strata.

[0016] The graph neural network model unit is used to classify risk coupling types into homogeneous factor coupling and heterogeneous factor coupling, and different sub-network models are used for feature extraction and coupling strength calculation respectively.

[0017] The dominant hazard identification unit is used to calculate the dominant hazard in the current environment by calling the coupling degree measurement model based on the hazard index and mutual coupling coefficient of each hazard output by the graph neural network.

[0018] In one or more embodiments of the present invention, the coupling degree measurement model is constructed based on the NK model or the system dynamics model and is used to quantify the risk coupling degree of a multi-hazard coupled system.

[0019] In one or more embodiments of the present invention, the virtual space decoupling deduction module includes:

[0020] The decoupling scheme generation unit is used to automatically match and generate at least one virtual decoupling treatment scheme, including pressure relief mining, gas drainage or dewatering measures, based on the dominant disaster element type.

[0021] A virtual simulation engine is used to simulate and extrapolate the changes in disaster parameter fields after the implementation of the virtual decoupled governance scheme in the 4D dynamic twin using the coupled nonlinear finite material point method.

[0022] The effect evaluation unit is used to compare the changes in the risk index and coupling strength coefficient of each disaster element before and after the simulation, to quantitatively evaluate the chain breaking effect of the virtual decoupling governance scheme, and to optimize and rank multiple schemes.

[0023] In one or more embodiments of the present invention, the system further includes a visualization and decision support module, which is used to dynamically display the 4D dynamic twin, the distribution area of ​​the dominant disaster element, and the evolution path of the disaster chain on a three-dimensional GIS map in the form of a "single map", and generate a decision report containing the preferred ranking of governance measures.

[0024] In one or more embodiments of the present invention, the real-time monitoring data of the Internet of Things includes at least one of microseismic monitoring data, ground stress data, gas concentration data, hydrological monitoring data, and temperature data; the mining progress data includes coal mining machine location information and roadway excavation footage information.

[0025] In one or more embodiments of the present invention, in the multidimensional dynamic twin construction module, for a geological model containing 10 million units, the time for a single dynamic update is less than 10 seconds.

[0026] In one or more embodiments of the present invention, in the virtual simulation engine, for a geological model containing 10 million units, the simulation time for a single virtual decoupling simulation is less than 45 seconds.

[0027] In one or more embodiments of the present invention, the following steps are included:

[0028] Step S1: Collect geological exploration data, IoT real-time monitoring data, and production system data to construct a three-dimensional static geological model;

[0029] Step S2: Based on real-time monitoring data and mining progress data, dynamically update the three-dimensional static geological model to form a 4D dynamic twin that evolves in real time as mining progresses;

[0030] Step S3: Construct a disaster coupling relationship graph model based on the mining space grid unit, quantify the coupling strength coefficient and disaster contribution of each disaster element through graph neural network, and identify the dominant disaster element and secondary disaster chain;

[0031] Step S4: Simulate the implementation effect of the virtual decoupling governance scheme in the 4D dynamic twin and output the chain break governance suggestion;

[0032] Step S5: Display the identified dominant disaster elements and chain disruption mitigation recommendations through a 3D visualization interface.

[0033] Compared with existing technologies, the beneficial effects of this invention are: it breaks through the limitations of static three-dimensional models, realizes the dynamic assimilation and on-the-fly updating of geological models and real-time monitoring data, and significantly improves the identification rate of hidden disaster-causing structures; it is the first to introduce graph neural networks into multi-hazard coupling analysis, realizing the quantitative calculation of the coupling relationship of multiple hazard elements such as gas, ground pressure, and water hazards, and the identification of dominant hazard elements, solving the problem of difficulty in quantifying the coupling effect of multiple hazards; it innovatively combines the decoupling concept with digital twin simulation, enabling virtual effect evaluation of governance schemes, transforming post-event handling into pre-event simulation, and providing a scientific basis for governance decisions; by eliminating coupling interference through decoupling analysis, the accuracy rate of single-hazard early warning can be improved to over 95%, and the overall reliability of the system reaches the international advanced level; it can achieve the integrated purpose of multi-hazard coupling decoupling analysis, dominant hazard element identification, and virtual governance simulation. Attached Figure Description

[0034] Figure 1 This is a system architecture diagram of the present invention;

[0035] Figure 2 This is a schematic diagram of the structure of the multidimensional dynamic twin construction module of the present invention;

[0036] Figure 3 This is a schematic diagram illustrating the construction of the disaster coupling relationship graph model of the present invention;

[0037] Figure 4 This is a flowchart illustrating the virtual decoupling derivation process of the present invention. Detailed Implementation

[0038] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0039] Example 1: The multi-hazard intelligent safety control system for coal mines in this example was applied to the 1305 working face of a high-gas outburst mine. This working face mainly mines the No. 2-1 coal seam, with a thickness of 4.5-6.2m, an average of 5.3m, and a burial depth of 680-750m. This mine faces the dual threat of gas outburst and rock burst, and has historically experienced multiple instances of abnormal gas outbursts and simultaneous increases in mine pressure manifestations. Traditional monitoring systems struggle to quantify the coupling relationship between these two factors.

[0040] Data acquisition layer deployment: Geological exploration data is imported from 3D seismic exploration results and data from 156 boreholes, including coal seam floor contour lines, fault attitude, and roof and floor lithology; IoT sensing layer deploys 86 gas sensors, including 20 at the working face, 18 in the return airway, 16 in the intake airway, 32 in the extraction pipeline, and 32 microseismic monitoring probes, covering the working face and a range of 200m before and after. 20 geostress gauges are buried on both sides of the roadways and the coal wall of the working face to collect real-time data on gas concentration, spatiotemporal distribution of microseismic events, and changes in geostress; The production system accesses the coal mining machine's position and advance data, automatically reads the coal mining machine's operation records daily, and obtains accurate mining progress.

[0041] Multidimensional dynamic twin construction: An initial three-dimensional geological model is constructed based on geological exploration data. The model covers the 1305 working face and its surrounding 500m area, including information such as coal seams, fault F13, a drop of 3.5m, and roof sandstone aquifers. The total number of model grid cells is approximately 8.2 million. At 2:00 AM every day, the system automatically triggers the dynamic update engine based on the previous day's mining progress and newly added monitoring data. The DSI interpolation algorithm is used to assimilate the newly revealed geological information and monitoring data into the model. The measured model update speed is 7.2 seconds for the 8.2 million cell model, which meets the real-time requirements of the field.

[0042] Construction of the disaster coupling relationship diagram: Taking the mining space within 200m around the mining face as the object, it is divided into 5m×5m×5m grid units, and a total of 1560 nodes are constructed. The average gas concentration, ground stress value and microseismic event frequency in each grid unit are used as node attributes. Connection edges between nodes are established according to the spatial distance of less than 10m or the existence of rock strata contact relationship, and a total of 8720 edges are established to form a disaster coupling relationship diagram model.

[0043] Graph Neural Network Model Training: The graph neural network model used in this embodiment is based on a graph attention network architecture, containing 3 graph convolutional layers and 2 fully connected layers. The model is pre-trained using monitoring data from the mine's history over the past 3 years. The training dataset includes monitoring data from 126 gas anomaly events and 43 mine pressure manifestation events. Risk coupling types are divided into two categories: homogeneous factor coupling and heterogeneous factor coupling. Different sub-network models are used for feature extraction for each category. Homogeneous factor coupling includes gas concentration and gas pressure, while heterogeneous factor coupling includes gas concentration and ground stress. The model training uses the Adam optimizer with an initial learning rate of 0.001, 200 training epochs, and an early stopping mechanism with a patience of 20.

[0044] Dominant disaster element identification: The real-time constructed disaster coupling relationship graph is input into the trained graph neural network model for inference calculation; the model output results show that the coupling strength coefficient between gas and ground pressure in the current area reaches 0.78, the coupling strength coefficient ranges from 0 to 1, and >0.7 indicates strong coupling; the gas dominant disaster element index is 0.65, the ground pressure dominant disaster element index is 0.42, the hydrological dominant disaster element index is 0.08, and the temperature dominant disaster element index is 0.05; based on this, the system identifies the disaster chain of "gas outburst as the dominant disaster element and rockburst as the associated secondary disaster", and highlights the distribution area of ​​the dominant disaster element in red in the 3D visualization interface, and shows the disaster chain propagation path from the gas-rich area to the ground pressure anomaly area with dynamic orange arrows.

[0045] To verify the accuracy of the identification results, on-site technicians conducted a specific verification of the identified area. In the identified dominant disaster-prone area, three verification boreholes were drilled in the 120-150m section of the working face machine roadway. The measured gas content was 8.6-9.2 m³ / t, which is 26%-35% higher than the average gas content of 6.8 m³ / t at the working face. At the same time, the daily frequency of microseismic events in this area increased from an average of 5 times / day to 16 times / day, confirming the accuracy of the model identification.

[0046] Decoupling treatment scheme generation: The system automatically matches the decoupling scheme library based on the identified dominant disaster element and generates three virtual decoupling schemes: (1) Gas pre-drainage scheme of this coal seam, drilling along the seam in the machine roadway and ventilation roadway of the working face, with a hole depth of 150m and a spacing of 10m; (2) Pressure relief blasting scheme, implementing deep hole pressure relief blasting in the 120-150m section of the machine roadway, with a blasting hole depth of 20m and a spacing of 5m; (3) Combined scheme, implementing gas pre-drainage and pressure relief blasting at the same time.

[0047] Virtual simulation: The coupled nonlinear finite material point method is used to simulate and simulate the three schemes in a 4D dynamic twin. Simulation parameter settings: time step 0.1s, total simulation duration simulates the mining progress in the next 30 days. During the simulation, the system calculates the gas pressure change, ground stress distribution adjustment and the evolution of the coupling relationship between each grid cell in real time.

[0048] Effect Evaluation: After the simulation is completed, the effect evaluation unit outputs the quantitative evaluation results of each scheme:

[0049] Option 1, gas pre-drainage: After 30 days of implementation, the gas pressure in the main disaster-causing area decreased from 0.85 MPa to 0.42 MPa, a decrease of 50.6%; the change in ground stress was not significant, and the coupling strength coefficient decreased from 0.78 to 0.52;

[0050] Option 2, pressure relief blasting: After implementation, the peak ground stress decreased from 32.6 MPa to 24.8 MPa, a decrease of 23.9%; the gas pressure decreased slightly from 0.85 MPa to 0.76 MPa, a decrease of 10.6%; and the coupling strength coefficient decreased to 0.61.

[0051] Option 3, combined scheme: After implementation, the gas pressure will drop to 0.41 MPa (a decrease of 51.8%), the peak ground stress will drop to 25.1 MPa (a decrease of 23.0%), and the coupling strength coefficient will drop to 0.31.

[0052] Based on the simulation results, the system output recommendations for fault-related mitigation: prioritizing a combined mitigation scheme of "primarily pre-draining gas from the coal seam, supplemented by pressure-relief blasting near the fault," and providing specific parameter suggestions for borehole construction. After implementing the mitigation scheme according to the system's recommendations on-site, no abnormal gas outbursts occurred during subsequent mining in the area, and the frequency of microseismic events returned to normal levels, verifying the effectiveness of the virtual simulation.

[0053] Example 2: This example was applied to the 3110 working face of a certain mine. The mine mainly mines the No. 3 coal seam, which is 3.2-4.5m thick, with an average of 3.8m, and a burial depth of 520-580m. During the mining of the working face, it faces the dual threats of a very thick aquifer and a hard roof. The aquifer is a 65m thick K8 sandstone aquifer, and the tensile strength of the roof is 8.6MPa. Historically, there have been combined disasters of roof water inrush accompanied by rock bursts.

[0054] The system deployment method is similar to that of Example 1; the three-dimensional geological model covers the 3110 working face and surrounding area, with a total of approximately 7.5 million model grid cells. In particular, a detailed model of the top aquifer was carried out, subdividing the K8 sandstone aquifer into 3 sub-layers, and importing dynamic water level data from 18 long-term hydrological wells.

[0055] The disaster coupling relationship diagram was constructed within a 200m range above the working face roof, and the node attributes included water pressure, in-situ stress, and rock fracture degree. The results of the graph neural network model inference calculation showed that the coupling strength coefficient between roof water pressure and in-situ stress reached 0.71, the dominant disaster element was identified as "roof water pressure", and the associated secondary disaster was "rockburst". Specific analysis showed that when the working face pushed past the vicinity of fault F5, the roof water pressure abnormally increased from 2.8MPa to 3.6MPa, and at the same time, microseismic events gathered near the fault zone, and the two showed obvious spatiotemporal correlation.

[0056] The system generates three virtual decoupling schemes: (1) Roof dewatering scheme, drilling holes for dewatering into the roof in the roadway, with a hole depth of 120m and a spacing of 20m; (2) Roof weakening scheme, using directional long borehole segmented fracturing technology to weaken the hard roof; (3) Combined scheme, simultaneously implementing dewatering and roof weakening.

[0057] The virtual simulation engine uses the coupled nonlinear finite matter point method for simulation. The simulation results show:

[0058] Option 1: After implementation, the roof water pressure will decrease from 3.6MPa to 1.8MPa, a decrease of 50%, and the change in ground stress will not be significant. The initial pressure step distance of the roof is expected to be 38m.

[0059] Option 2: After implementation, the roof weakening effect is significant, and the expected pressure step distance is extended from 42m to 52m, an increase of 23.8%, but the water pressure does not change significantly;

[0060] Combined solution: After implementation, the water pressure dropped to 1.9 MPa, a decrease of 47.2%, the pressure step distance was extended to 51m, an increase of 21.4%, and the coupling strength coefficient decreased from 0.71 to 0.35.

[0061] The system output recommended prioritizing a combined treatment approach of "primarily dewatering the roof, supplemented by weakening the roof near the fault." Following this recommendation, 16 drainage boreholes were drilled on-site, releasing a total of 86,000 m³ of water and reducing the water pressure to below 2.0 MPa. Simultaneously, segmented fracturing was implemented in the fault-affected area. During actual mining, the roof pressure step distance was 48 m, and no water inrush or rockburst events occurred.

[0062] Example 3: This example demonstrates the overall system deployment in a large coal mining group to verify the system's performance and reliability in a real production environment.

[0063] Hardware deployment architecture

[0064] The system adopts a three-layer architecture of "cloud-edge-device":

[0065] On the end side: a total of 1,268 sensors of various types are deployed, including 326 gas sensors, 128 microseismic monitoring probes, 86 ground stress gauges, 52 hydrological monitoring instruments, 76 temperature sensors, etc.; 600 smart safety helmets are provided, integrating UWB positioning modules and voice interaction functions.

[0066] Side-side: 16 edge computing nodes are deployed in the main underground roadways and mining faces. Each node is equipped with an Intel Xeon 4214 processor (12 cores and 24 threads), 64GB of memory, and 2TB of SSD storage. They run edge computing service software to achieve real-time data preprocessing and local risk field construction.

[0067] Cloud side: The ground data center deploys a server cluster, including 4 compute nodes (each with 2 Intel Xeon Gold 6248 processors and 512GB of memory), 2 GPU servers (each with 4 NVIDIA V100 graphics cards) for graph neural network model training and virtual inference calculations, and 2 storage servers (total capacity 1.2PB).

[0068] Model update performance test: Dynamic update tests were conducted on 3D geological models of different sizes. The test results are as follows:

[0069] 200 2.3 2.8 yes 500 4.6 5.3 yes 800 7.1 8.2 yes 1000 9.2 10.8 Yes (<10 seconds) 1200 11.5 13.6 Partial timeout

[0070] Test results show that for geological models with up to 10 million units, the system can complete dynamic updates within 10 seconds, meeting the real-time requirements of the field.

[0071] Virtual simulation performance test: Virtual simulation tests were conducted using geological models of different sizes. The test results are as follows:

[0072] 200 9.8 11.2 Yes (<45 seconds) 500 21.3 24.5 yes 800 35.6 39.8 yes 1000 42.1 46.3 Yes (average over 45 seconds) 1200 58.4 65.7 no

[0073] Test results show that for geological models with up to 10 million units, the system can complete a single virtual simulation within 45 seconds, meeting the timeliness requirements for governance decision-making.

[0074] Graph Neural Network Inference Performance Test: The inference speed of the graph neural network model was tested on a GPU server. For a disaster coupling graph containing 1500 nodes and 8000 edges, the single inference time was 0.12 seconds, which meets the requirements for real-time monitoring.

[0075] After the system had been running for six months, a statistical analysis of the early warning accuracy was conducted. Based on actual or non-occurring disaster events on-site, the statistical results are as follows:

[0076] gas 86 82 4 2 95.3% rock burst 43 40 3 1 93.0% roof 52 48 4 2 92.3% Hydrology 28 26 2 1 92.9% comprehensive 209 196 13 6 93.8%

[0077] The overall early warning accuracy rate reached 93.8%, which is higher than the existing KJ90X system's 85%-8 and the similar system's 90%-5, ​​verifying the technical effect of the present invention in eliminating coupling interference and improving early warning accuracy through decoupling analysis.

[0078] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0079] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A multi-hazard intelligent safety control system for underground coal mines, characterized in that, include: The multidimensional dynamic twin construction module is used to construct a three-dimensional static geological model based on geological exploration data, and dynamically update the three-dimensional static geological model based on real-time monitoring data from the underground Internet of Things and mining progress data, forming a 4D dynamic twin that evolves in real time as the mining face advances; The multi-hazard coupling and decoupling analysis module based on graph neural networks is used to construct a disaster coupling relationship graph model with mining space grid cells as nodes, monitoring data of multiple disasters as node attributes, and spatial proximity and physical interaction relationships between nodes as edges. The disaster coupling relationship graph model is inferred and calculated by a pre-trained graph neural network to quantify the coupling strength coefficient and disaster contribution of each disaster element, and the dominant disaster element and its associated secondary disaster chain in the current mining environment are identified based on the calculation results. The virtual space decoupling simulation module is used to generate at least one virtual decoupling governance scheme for the identified dominant disaster element, and to simulate the disaster evolution process after implementing the virtual decoupling governance scheme in the 4D dynamic twin, and output chain break suggestions to guide on-site governance.

2. The intelligent safety control system for multiple disasters in coal mines according to claim 1, characterized in that, The multidimensional dynamic twin construction module includes: The static geological modeling unit is used to construct a three-dimensional static geological model containing information on coal seams, faults, collapse columns, and aquifers based on borehole data, seismic exploration data, and well logging data. The real-time data assimilation unit is used to access the real-time monitoring data of the Internet of Things and use a discrete smooth interpolation algorithm or a point-layer thickness grid model that integrates vector and grid to assimilate the mining exposure information and monitoring data into the three-dimensional static geological model. The dynamic update engine is used to trigger local updates of the model based on mining progress data, so as to realize the geological model and production progress in line with mining.

3. The intelligent safety control system for multiple disasters in coal mines according to claim 1, characterized in that, The multi-hazard coupling decoupling analysis module based on graph neural networks includes: The disaster coupling relationship graph construction unit is used to divide the mining space into multiple grid units as nodes of the graph model, and to use one or more monitoring data of gas concentration, ground stress value, water pressure, and temperature as node attributes. At the same time, the connection edges between nodes are established according to spatial distance and physical and mechanical properties of rock strata. The graph neural network model unit is used to classify risk coupling types into homogeneous factor coupling and heterogeneous factor coupling, and different sub-network models are used for feature extraction and coupling strength calculation respectively. The dominant hazard identification unit is used to calculate the dominant hazard in the current environment by calling the coupling degree measurement model based on the hazard index and mutual coupling coefficient of each hazard output by the graph neural network.

4. The intelligent safety control system for multiple disasters in coal mines according to claim 3, characterized in that, The coupling degree measurement model is constructed based on the NK model or system dynamics model and is used to quantify the risk coupling degree of multi-hazard coupled systems.

5. The intelligent safety control system for multiple disasters in coal mines according to claim 1, characterized in that, The virtual space decoupling deduction module includes: The decoupling scheme generation unit is used to automatically match and generate at least one virtual decoupling treatment scheme, including pressure relief mining, gas drainage or dewatering measures, based on the dominant disaster element type. A virtual simulation engine is used to simulate and extrapolate the changes in disaster parameter fields after the implementation of the virtual decoupled governance scheme in the 4D dynamic twin using the coupled nonlinear finite material point method. The effect evaluation unit is used to compare the changes in the risk index and coupling strength coefficient of each disaster element before and after the simulation, to quantitatively evaluate the chain breaking effect of the virtual decoupling governance scheme, and to optimize and rank multiple schemes.

6. The intelligent safety control system for multiple disasters in coal mines according to claim 1, characterized in that, The system also includes a visualization and decision support module, which dynamically displays the 4D dynamic twin, the distribution area of ​​the dominant disaster element, and the evolution path of the disaster chain on a 3D GIS map in the form of a "single map", and generates a decision report containing the preferred ranking of governance measures.

7. The intelligent safety control system for multiple disasters in coal mines according to claim 1, characterized in that, The IoT real-time monitoring data includes at least one of microseismic monitoring data, ground stress data, gas concentration data, hydrological monitoring data, and temperature data; the mining progress data includes coal mining machine location information and roadway excavation progress information.

8. The intelligent safety control system for multiple disasters in coal mines according to claim 1, characterized in that, In the multidimensional dynamic twin construction module, for a geological model containing 10 million units, the time for a single dynamic update is less than 10 seconds.

9. The intelligent safety control system for multiple disasters in coal mines according to claim 5, characterized in that, In the virtual simulation engine, for a geological model containing 10 million elements, the simulation time for a single virtual decoupling simulation is less than 45 seconds.

10. A method for intelligent safety control of multiple hazards in coal mines based on the system described in any one of claims 1-9, characterized in that, Includes the following steps: Step S1: Collect geological exploration data, IoT real-time monitoring data, and production system data to construct a three-dimensional static geological model; Step S2: Based on real-time monitoring data and mining progress data, dynamically update the three-dimensional static geological model to form a 4D dynamic twin that evolves in real time as mining progresses; Step S3: Construct a disaster coupling relationship graph model based on the mining space grid unit, quantify the coupling strength coefficient and disaster contribution of each disaster element through graph neural network, and identify the dominant disaster element and secondary disaster chain; Step S4: Simulate the implementation effect of the virtual decoupling governance scheme in the 4D dynamic twin and output the chain break governance suggestion; Step S5: Display the identified dominant disaster elements and chain disruption mitigation recommendations through a 3D visualization interface.

Images