Mine safety risk early warning method and system based on Internet of Things

Abstract

The invention discloses a mine safety risk early warning method and system based on the Internet of Things, and the method comprises the steps: obtaining historical environment safety data, constructing a mine safety map neural network and a mine safety prediction model, determining a mine monitoring objective function, optimizing network parameters, obtaining an optimized perception network, obtaining internal perception data and external perception data, and carrying out the early warning of the mine safety risk. Obtaining an external influence result and an internal and external association result from the external sensing data, performing first-class mine external early warning according to the external influence result, correcting the internal sensing data to obtain corrected internal sensing data, screening the corrected internal sensing data by adopting a dynamic multi-sphere, and performing second-class mine internal early warning according to a screening result. And matching the results of the first-class mine external early warning and the second-class mine internal early warning to carry out third-class mine safety early warning. The method not only can improve the accuracy of mine safety risk early warning, but also has good interpretability, and can be directly applied to a mine safety risk early warning system.

Description

Technical Field

[0001] The present invention relates to the field of safety early warning, and in particular to a mine safety risk early warning method and system based on the Internet of Things. Background Art

[0002] With the development of Internet of Things technology, mine safety production is undergoing profound changes. As an important place for resource extraction, mines have complex production and operation environments and are subject to various safety risks such as gas explosions, water seepage, and roof collapse. Using Internet of Things technology to build an intelligent safety risk early warning system to achieve real-time monitoring, accurate analysis, and early warning of the mine environment and operating status has become a key means to ensure mine safety production, reduce the incidence of accidents, and reduce casualties and economic losses.

[0003] Traditional mine safety risk warning technologies often rely on local data collection from a single type of sensor, resulting in high data transmission latency and poor stability, and a lack of in-depth mining capabilities for complex correlations between data. This results in insufficient accuracy and timeliness of warnings, making it difficult to effectively respond to complex and changing mine safety risks. In recent years, emerging intelligent perception and data analysis technologies have provided new development directions for mine safety warnings. The present invention designs a mine safety risk warning method and system based on the Internet of Things. By constructing a mine safety graph neural network and a mine safety prediction model, combined with an optimized perception network design, and comprehensively utilizing 5G private networks and wide area networks to achieve efficient transmission and analysis of internal and external data, the corrected internal perception data is processed through a dynamic multi-sphere screening algorithm, which can accurately identify safety risks within the mine and conduct comprehensive warnings based on external influencing factors. This overcomes the shortcomings of traditional technologies, improves the comprehensiveness, accuracy, and real-time nature of mine safety risk warnings, provides a more reliable technical guarantee for mine safety production, and has important practical value in promoting the intelligent upgrade of the mining industry. Summary of the Invention

[0004] The purpose of the present invention is to provide a mine safety risk early warning method and system based on the Internet of Things.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Acquire historical environmental safety data, and construct a mine safety graph neural network and a mine safety prediction model based on the historical environmental safety data;

[0008] Determining a mine monitoring objective function, setting an initial perception network and optimizing network parameters according to the mine monitoring objective function to obtain an optimized perception network; the perception network includes an internal perception network and an external perception network; the network parameters include sensor locations and the number of sensors;

[0009] Acquiring internal perception data and external perception data according to the optimized perception network, and inputting the external perception data into the mine safety graph neural network and the mine safety prediction model to obtain external impact results and internal and external correlation results; the internal perception data is transmitted via a 5G private network; and the external perception data is transmitted via a wide area network;

[0010] Performing a first-class mine external warning based on the external impact result, correcting the internal perception data based on the external impact result and the internal-external correlation result to obtain corrected internal perception data, using a dynamic multi-sphere to filter the corrected internal perception data, and performing a second-class mine internal warning based on the filtering result;

[0011] Class III mine safety warnings are issued based on the external warning results of Class I mines and the internal warning results of Class II mines.

[0012] Furthermore, the method for constructing a mine safety graph neural network and a mine safety prediction model includes:

[0013] Acquire historical environmental safety data via a low-power wide area network, including mine structural parameters, meteorological data, geological data, activity data, historical ventilation and drainage and equipment anomaly records, historical rock formation damage records, and historical support structure deformation records;

[0014] Perform spatiotemporal alignment of historical environmental safety data, linking meteorological data with historical ventilation, drainage, and equipment anomaly records, geological data with historical rock formation damage records, and activity data with historical support structure deformation records;

[0015] Constructing a mine safety graph neural network, which includes a dynamic graph structure, a graph attention network, and a spatiotemporal graph convolution module;

[0016] The nodes of the dynamic graph structure include spatial region nodes and data entity nodes, and the edge relationships include spatial affiliation edges, causal influence edges, and risk transmission edges. The graph attention network is used to aggregate node features and perform cross-modal feature interaction on meteorological, geological, and activity data. The spatiotemporal graph convolution module extracts dynamic risk propagation patterns and outputs the risk category and probability of each node.

[0017] Cross entropy loss and KL divergence are used as loss functions to constrain the consistency between the predicted risk distribution and historical disaster records, and the associated historical environmental safety data are used to train and evaluate the mine safety graph neural network.

[0018] Constructing a mine safety prediction model, wherein the mine safety prediction model includes a roof pressure prediction base model and a groundwater level prediction base model;

[0019] The roof pressure prediction base model receives mine structural parameters, geological data, activity data and corresponding associated data, and uses a dual-channel LSTM-BP hybrid structure to extract temporal features and static features to predict the distribution of roof pressure inside the mine;

[0020] The groundwater level prediction base model receives mine structural parameters, meteorological data and corresponding associated data, uses the Wavelet-BP network to separate the trend term and periodic term of water level change through wavelet decomposition, and predicts the water level increase and water inrush risk level at different mining depths.

[0021] Furthermore, the method for obtaining an optimized perception network includes:

[0022] The internal sensing network places multi-source sensors inside the mine to acquire internal sensing data; the internal sensing data includes gas concentration, roof pressure, groundwater level, equipment data, and mine tunnel images; the external sensing network places multi-source sensors outside the mine to acquire external sensing data; the external sensing data includes meteorological data, geological data, and activity data;

[0023] The objective function of mine monitoring is determined by maximizing the monitoring range, minimizing the layout and operation costs, and minimizing the adjustment range. The expression is:

[0024]

[0025] Where Aim is the mine monitoring objective function, α is the monitoring coverage weight, A is the detection coverage, β is the cost weight, Cost is the monitoring cost, λ1 is the adjustment penalty coefficient, Δ adjust For adjustment range, M=M in +M out is the number of sensor categories in the perception network, which is equal to the number of sensor categories in the internal perception network M in External sensing network sensor category M out The sum, w r is the perception network weight, R = {in, out} is the perception network set, w in is the internal perception network weight, w out is the external perception network weight, w m is the weight of m type sensor, N m is the number of sensors of type m, A m is the monitoring range of type m sensor, A m,stais the rated monitoring range of m-type monitoring data, Overlap(·) is the penalty term for the overlapping area of the internal and external networks, and A in is the coverage area of the internal sensor network, A out is the coverage area of the external sensor network, t i 、 are respectively the deployment cost, energy cost per unit time, running time and number of type i sensors in the internal perception network, t j 、 are respectively the deployment cost, energy cost per unit time, running time and number of type i sensors in the external perception network, is the Euclidean distance of the position change of m types of sensors, To adjust the position of the m-type sensor, To adjust the position of the front m type sensor, is the number of m type sensors after adjustment, To adjust the number of sensors in the first m categories;

[0026] Set up the initial perception network, use the network parameters of the initial perception network as the initial population, and use the particle swarm-simulated annealing algorithm to optimize the network parameters to obtain the optimal network parameters. The specific steps are as follows:

[0027] Initialize the population and update the particle speed and position. The expression is:

[0028]

[0029] in is the velocity update of particle i in dimension d at the k+1 iteration, is the speed corresponding to k iterations, is the position update of particle i in dimension d at k+1 iterations, is the position corresponding to k iterations, ω0 is the inertia weight, T k+1 、T k are the annealing temperatures for k+1 iterations and k iterations respectively, c1 and c2 are learning factors, r1 and r2 are random numbers in the range of (0,1), is the individual extreme value of particle i in dimension d in k iterations, that is, the individual optimal position of k iterations, is the global extreme value of the population in dimension d in k iterations, that is, the optimal position of the population in k iterations, ξ k+1 ,ξ k are the adaptive perturbation intensities for k+1 iterations and k iterations respectively, is a Gaussian random disturbance with variance σ 2 =T k , T0 is the initial temperature, v tis the basic cooling rate, is the particle swarm position variance measure for k iterations, is the number of particles in the particle swarm, which is equal to the number of sensors, P accept is the acceptance probability, Aim(X new )、Aim(X old ) are the objective functions of the new population and the old population respectively;

[0030] The dual Metropolis criterion is used to update individual extreme values and global extreme values simultaneously. The expression is:

[0031]

[0032] in are the individual extreme value and global extreme value updated by k+1 iterations respectively, is the objective function of the position corresponding to the k+1 iteration of individual i in dimension d, are the objective functions of the individual extreme values of individual i in dimension d at k iterations and k+1 iterations, respectively. is the objective function value of the global extreme value of the population in dimension d at k iterations;

[0033] Determine the network parameters based on the updated population position and calculate the mine monitoring objective function. Repeat the iteration until the mine monitoring objective function is minimized and output the optimal network parameters. Set them according to the optimal network parameters to obtain the optimized perception network.

[0034] A time threshold and a project difference threshold are set. When any of the thresholds is met, the network optimization operation is triggered and the network parameters are updated. The project difference threshold represents the difference between the current mine project progress and the mine project progress at the last network parameter update time.

[0035] Furthermore, the method for performing a type of mine external early warning includes:

[0036] Optimize the external perception network through low-power wide area network connection to transmit external perception data to the edge end, and optimize the internal perception network through 5G private network connection to transmit internal perception data to the cloud platform;

[0037] Performing edge computing on external perception data at the edge to obtain edge processing results, specifically including: inputting the external perception data into the mine safety graph neural network to obtain external impact results, and inputting the external perception data into the mine safety prediction model to obtain internal and external correlation results; the edge processing results include external impact results and internal and external correlation results; the external impact results include risk categories and risk probabilities; the internal and external correlation results include predicted values for roof pressure distribution and groundwater level predictions;

[0038] A type of mine external warning is performed at the edge end based on the external impact results; the type of mine external warning includes: an external low-risk warning when the risk probability is greater than the first probability threshold, an external medium-risk warning when the risk probability is greater than the second probability threshold, and an external high-risk warning when the risk probability is greater than the third probability threshold.

[0039] Furthermore, the method for obtaining and correcting internal perception data includes:

[0040] The edge processing results are transmitted to the cloud platform via the 5G private network. The roof pressure distribution prediction value of the internal and external correlation results is fused with the roof pressure of the internal sensing data to obtain the corrected roof pressure. The groundwater level prediction value of the internal and external correlation results is fused with the groundwater level of the internal sensing data to obtain the corrected groundwater level.

[0041] Extract the ventilation risk probability from the external impact results, calculate the deviation of gas concentration influencing factors based on external and internal sensing data, determine the ventilation factor based on the deviation of gas concentration influencing factors, and determine the corrected gas concentration based on the ventilation factor, ventilation risk probability, and gas concentration. The deviation of gas concentration influencing factors includes internal and external temperature deviation, internal and external humidity deviation, internal and external air pressure deviation, and internal and external wind speed deviation. The ventilation factor expression is:

[0042]

[0043] Among them F v is the ventilation factor, δ1 is the temperature and humidity weight, δ2 is the airflow weight, ΔT is the internal and external temperature deviation, ΔH is the internal and external humidity deviation, ΔP is the internal and external pressure deviation, P0 is the pressure reference value, ΔV is the internal and external wind speed deviation, V0 is the wind speed reference value;

[0044] Extract the equipment risk probability from the external impact results, divide the equipment data of the internal perception data into equipment operation data and equipment status data, determine the equipment status factor based on the equipment status data, and determine the modified equipment operation data based on the equipment status factor, equipment risk probability and equipment operation data; the equipment status factor expression is:

[0045]

[0046] Among them F m is the device status factor, Q is the number of devices, μ l is the maintenance weight of equipment l, is the location weight of equipment l in the mine tunnel, N l is the number of maintenance times of equipment l in one year, N l,0 is the industry benchmark annual maintenance frequency of equipment l, t l The duration of the most recent single maintenance of equipment l. is the average repair time of device l, τ l is the material degradation coefficient of equipment l, t l,tol is the cumulative running time of device l, t l,con is the continuous operation time of device l, t l,des is the designed operating time of device 1;

[0047] Performing image processing on the mine tunnel image to obtain mine tunnel image features; the mine tunnel image features include the number of mine tunnel cracks, the number and width of mine tunnel cracks, roof deformation value, equipment deformation value and foreign matter conditions;

[0048] The corrected roof pressure, corrected groundwater level, corrected gas concentration, corrected equipment operation data and mine tunnel image features are combined to form the corrected internal perception data.

[0049] Furthermore, the method for performing internal early warning of a Class II mine includes:

[0050] Dynamic multi-sphere screening is used to correct internal perception data. The corrected internal perception data outside the multi-sphere is defined as internal risk warning data. The ratio of different types of internal risk warning data to the corresponding sphere radius is calculated to obtain different types of internal risk degrees. The radius update expression of the multi-sphere is:

[0051]

[0052] Where r(t) is the updated sphere radius at time t, r0 is the initial sphere radius, θ t is the time attenuation factor, λ2 is the density weight, λ3 is the abnormal data weight, ρ(t) is the data density of the sphere coverage area at the current moment, ∈ is the smoothing coefficient, N ano (t) is the count of abnormal data in the current sphere;

[0053] Two types of internal mine warnings are carried out in the cloud based on different types of internal risk levels; the two types of internal mine warnings include: internal low-risk warnings when the risk level is greater than the first risk level threshold, internal medium-risk warnings when the risk level is greater than the second risk level threshold, and internal high-risk warnings when the risk level is greater than the third risk level threshold.

[0054] Furthermore, the method for performing three types of mine safety early warning includes:

[0055] The first-class mine external warning results are transmitted to the cloud platform via the 5G private network. The grid area where the first-class mine external warning results are located is matched with the sensor location corresponding to the second-class mine internal warning results. Three types of mine safety warnings are performed on the cloud for the same grid area. The first-class mine external warning and the second-class mine internal warning do not affect each other. The three types of mine safety warnings include mine safety low-risk warning, mine safety medium-risk warning, mine safety high-risk warning and mine safety emergency risk warning.

[0056] The strategies for the three types of mine safety warnings are as follows: when the risk levels of the external warning of a type one mine and the internal warning of a type two mine are the same, the risk level of the mine safety warning is increased; when the risk levels of the external warning of a type one mine and the internal warning of a type two mine are different, the higher risk level of the two is selected as the mine safety warning risk level; when the risk level of any type of warning is high risk and the risk level of the other type of warning is low risk or above, the mine safety warning risk level is defined as an emergency risk.

[0057] Secondly, a mine safety risk early warning system based on the Internet of Things includes:

[0058] Network parameter module: used to determine the mine monitoring objective function, set the initial perception network and optimize the network parameters according to the mine monitoring objective function to obtain an optimized perception network, and obtain internal perception data and external perception data according to the optimized perception network;

[0059] Edge computing module: used to build a mine safety graph neural network and a mine safety prediction model based on historical environmental safety data, input the external perception data into the mine safety graph neural network and the mine safety prediction model to obtain external impact results and internal and external correlation results, and perform a type of mine external early warning based on the external impact results;

[0060] Cloud computing module: used to correct the internal perception data according to the external influence result and the internal and external correlation result to obtain corrected internal perception data, use dynamic multi-sphere to filter the corrected internal perception data, and perform Class II mine internal warning according to the filtering result, and perform Class III mine safety warning by matching the results of Class I mine external warning and Class II mine internal warning;

[0061] Management module: used to manage the results of the first-class mine external warning, the second-class mine internal warning and the third-class mine safety warning, and perform mine safety management according to the warning results.

[0062] The beneficial effects of the present invention are:

[0063] The present invention is a mine safety risk early warning method and system based on the Internet of Things. Compared with the existing technology, the present invention has the following technical effects:

[0064] The present invention can improve the accuracy of mine safety risk warning by constructing models, optimizing network parameters, data correction, multi-sphere screening and classification and grading warning steps, thereby improving the precision of mine safety risk warning. Optimizing mine safety risk warning technology can greatly save resources and improve work efficiency. It can realize risk warning of mine safety and conduct real-time data monitoring and protection warning of mines, which is helpful to reduce mine accident rate, ensure the safety of personnel and reduce property loss. It has far-reaching significance for promoting the sustainable development of the mining industry and maintaining social stability. It can adapt to different mine safety risk warning systems and the mine safety risk warning needs of different users, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flowchart of the steps of a mine safety risk early warning method based on the Internet of Things of the present invention. DETAILED DESCRIPTION

[0066] The present invention will be further described below through specific examples. The illustrative examples and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0067] The present invention provides a mine safety risk early warning method and system based on the Internet of Things, comprising the following steps:

[0068] like Figure 1 As shown, in this embodiment, the following steps are included:

[0069] Acquire historical environmental safety data, and construct a mine safety graph neural network and a mine safety prediction model based on the historical environmental safety data;

[0070] Determining a mine monitoring objective function, setting an initial perception network and optimizing network parameters according to the mine monitoring objective function to obtain an optimized perception network; the perception network includes an internal perception network and an external perception network; the network parameters include sensor locations and the number of sensors;

[0071] Acquiring internal perception data and external perception data according to the optimized perception network, and inputting the external perception data into the mine safety graph neural network and the mine safety prediction model to obtain external impact results and internal and external correlation results; the internal perception data is transmitted via a 5G private network; and the external perception data is transmitted via a wide area network;

[0072] Performing a first-class mine external warning based on the external impact result, correcting the internal perception data based on the external impact result and the internal-external correlation result to obtain corrected internal perception data, using a dynamic multi-sphere to filter the corrected internal perception data, and performing a second-class mine internal warning based on the filtering result;

[0073] Class III mine safety warnings are issued based on the external warning results of Class I mines and the internal warning results of Class II mines.

[0074] In this embodiment, the method for constructing a mine safety graph neural network and a mine safety prediction model includes:

[0075] Acquire historical environmental safety data via a low-power wide area network, including mine structural parameters, meteorological data, geological data, activity data, historical ventilation and drainage and equipment anomaly records, historical rock formation damage records, and historical support structure deformation records;

[0076] Perform spatiotemporal alignment of historical environmental safety data, linking meteorological data with historical ventilation, drainage, and equipment anomaly records, geological data with historical rock formation damage records, and activity data with historical support structure deformation records;

[0077] Constructing a mine safety graph neural network, which includes a dynamic graph structure, a graph attention network, and a spatiotemporal graph convolution module;

[0078] The nodes of the dynamic graph structure include spatial region nodes and data entity nodes, and the edge relationships include spatial affiliation edges, causal influence edges, and risk transmission edges. The graph attention network is used to aggregate node features and perform cross-modal feature interaction on meteorological, geological, and activity data. The spatiotemporal graph convolution module extracts dynamic risk propagation patterns and outputs the risk category and probability of each node.

[0079] Cross entropy loss and KL divergence are used as loss functions to constrain the consistency between the predicted risk distribution and historical disaster records, and the associated historical environmental safety data are used to train and evaluate the mine safety graph neural network.

[0080] Constructing a mine safety prediction model, wherein the mine safety prediction model includes a roof pressure prediction base model and a groundwater level prediction base model;

[0081] The roof pressure prediction base model receives mine structural parameters, geological data, activity data and corresponding associated data, and uses a dual-channel LSTM-BP hybrid structure to extract temporal features and static features to predict the distribution of roof pressure inside the mine;

[0082] The groundwater level prediction model receives mine structural parameters, meteorological data and corresponding related data, uses Wavelet-BP network to separate the trend term and periodic term of water level change through wavelet decomposition, and predicts the water level increase and water inrush risk level at different mining depths;

[0083] In the actual evaluation, after acquiring historical environmental safety data through a low-power wide area network (using NB-IoT technology), spatiotemporal alignment was performed;

[0084] Associate meteorological data with historical ventilation, drainage, and equipment anomaly records. For example, query and associate the ventilation and drainage system load during a certain period with a significant increase in precipitation; query and associate the heat dissipation burden and unstable equipment operation of the ventilation equipment when the temperature is too high; associate geological data with historical rock damage records. For example, query and associate the rock damage during mining when the rock hardness is low and there are fault structures; query and associate the damage and deformation inside the mine when earthquakes / geological subsidence occur; associate activity data with historical support structure deformation records. For example, query and associate the support structure deformation records when the mining volume / surrounding traffic volume increases significantly;

[0085] The H3 geographic grid coding technology is used to dynamically divide the physical space of the mine into multiple grid units with different granularities to obtain spatial regional nodes. For areas with frequent mining activities and complex geological conditions, fine grid division is used, while for relatively stable areas, coarser grids are used.

[0086] Monitoring equipment or historical risk events determine data entity nodes. Monitoring equipment nodes contain information such as equipment type, location, and monitoring data. Historical risk event nodes record detailed information such as the time, location, type, scope of impact, and related handling measures of the event.

[0087] Define spatial affiliation edges to connect data entity nodes with the spatial region nodes to which they belong. For example, a gas sensor located in a mining area is used as a data entity node and is connected to the spatial region node representing the mining area through a spatial affiliation edge.

[0088] The transfer entropy algorithm is used to identify cross-domain causal relationships between meteorological, geological, and activity data nodes and determine causal influence edges. For example, an increase in precipitation in a certain area will cause the groundwater level to rise, which in turn affects the rock stability in the area and increases the risk of rock formation damage. In this case, causal influence edges are established between the precipitation data node and the groundwater level data node, and between the groundwater level data node and the rock formation damage data node.

[0089] Based on the similarity matching of historical event characteristics, risk pattern migration paths are established to determine risk transmission edges. For example, when the intensity of mining activities in a certain area suddenly increases and the support structure undergoes slight deformation, a roof collapse accident occurs. A similar situation occurs in another area. The relevant data entity nodes of these two areas are connected to establish a risk transmission edge, indicating that the two areas may have similar risk pattern migration paths.

[0090] The following formula is used to calculate the cross entropy loss and KL divergence of the true risk distribution and the risk distribution predicted by the model:

[0091]

[0092] Where H(p,q) is the cross entropy loss between the true risk distribution p and the risk distribution q predicted by the model, and D KL (p||q) is the KL divergence between the true risk distribution p and the model-predicted risk distribution q, p(i) is the true probability of risk i, q(i) is the model-predicted distribution of risk i, and O is the number of risk types.

[0093] In this embodiment, the method for obtaining an optimized perception network includes:

[0094] The internal sensing network places multi-source sensors inside the mine to acquire internal sensing data; the internal sensing data includes gas concentration, roof pressure, groundwater level, equipment data, and mine tunnel images; the external sensing network places multi-source sensors outside the mine to acquire external sensing data; the external sensing data includes meteorological data, geological data, and activity data;

[0095] The objective function of mine monitoring is determined by maximizing the monitoring range, minimizing the layout and operation costs, and minimizing the adjustment range. The expression is:

[0096]

[0097] Where Aim is the mine monitoring objective function, α is the monitoring coverage weight, A is the detection coverage, β is the cost weight, Cost is the monitoring cost, λ1 is the adjustment penalty coefficient, Δ adjust For adjustment range, M=M in +M out is the number of sensor categories in the perception network, which is equal to the number of sensor categories in the internal perception network M in External sensing network sensor category M out The sum, w r is the perception network weight, R = {in, out} is the perception network set, w in is the internal perception network weight, w out is the external perception network weight, w m is the weight of m type sensor, N m is the number of sensors of type m, A m is the monitoring range of type m sensor, A m,sta is the rated monitoring range of m-type monitoring data, Overlap(·) is the penalty term for the overlapping area of the internal and external networks, and A in is the coverage area of the internal sensor network, A outis the coverage area of the external sensor network, t i 、 are respectively the deployment cost, energy cost per unit time, running time and number of type i sensors in the internal perception network, t j 、 are respectively the deployment cost, energy cost per unit time, running time and number of type i sensors in the external perception network, is the Euclidean distance of the position change of m types of sensors, To adjust the position of the m-type sensor, To adjust the position of the front m type sensor, is the number of m type sensors after adjustment, To adjust the number of sensors of the first m categories;

[0098] Set up the initial perception network, use the network parameters of the initial perception network as the initial population, and use the particle swarm-simulated annealing algorithm to optimize the network parameters to obtain the optimal network parameters. The specific steps are as follows:

[0099] Initialize the population and update the particle speed and position. The expression is:

[0100]

[0101] in is the velocity update of particle i in dimension d at the k+1 iteration, is the speed corresponding to k iterations, is the position update of particle i in dimension d at k+1 iterations, is the position corresponding to k iterations, ω0 is the inertia weight, T k+1 、T k are the annealing temperatures for k+1 iterations and k iterations respectively, c1 and c2 are learning factors, r1 and r2 are random numbers in the range of (0,1), is the individual extreme value of particle i in dimension d in k iterations, that is, the individual optimal position of k iterations, is the global extreme value of the population in dimension d in k iterations, that is, the optimal position of the population in k iterations, ξ k+1 ,ξ k are the adaptive perturbation intensities for k+1 iterations and k iterations respectively, is a Gaussian random perturbation with variance σ 2 =T k , T0 is the initial temperature, v t is the basic cooling rate, is the particle swarm position variance measure for k iterations, is the number of particles in the particle swarm, which is equal to the number of sensors, P acceptis the acceptance probability, Aim(X new )、Aim(X old ) are the objective functions of the new population and the old population respectively;

[0102] The dual Metropolis criterion is used to update individual extreme values and global extreme values simultaneously. The expression is:

[0103]

[0104] in are the individual extreme value and global extreme value updated by k+1 iterations respectively, is the objective function of the position corresponding to the k+1 iteration of individual i in dimension d, are the objective functions of the individual extreme values of individual i in dimension d at k iterations and k+1 iterations, respectively. is the objective function value of the global extreme value of the population in dimension d at k iterations;

[0105] Determine the network parameters based on the updated population position and calculate the mine monitoring objective function. Repeat the iteration until the mine monitoring objective function is minimized and output the optimal network parameters. Set them according to the optimal network parameters to obtain the optimized perception network.

[0106] Set a time threshold and a project difference threshold. When either threshold is met, the network optimization operation is triggered and the network parameters are updated. The project difference threshold represents the difference between the current mine project progress and the mine project progress at the time of the last network parameter update.

[0107] In the actual assessment, a risk warning is carried out for a mine, and the monitoring coverage weight α is 5, the cost weight β is 0.05, the adjustment penalty coefficient λ1 is 0.1, and the internal perception network weight w is in is 0.6, external perception network weight w out is 0.4, and the mine monitoring objective function of the initial sensor network is calculated based on the one-day operation of the sensing network, with a value of 20 (detection coverage A is 0.5, monitoring cost Cost is 2 million yuan, and adjustment range Δ adjust =0);

[0108] Take the inertia weight ω0 as 0.8, the learning factors c1 and c2 as 1.5, the initial temperature T0 as 500, and the basic cooling rate v t The particle swarm-simulated annealing algorithm is used to optimize the network parameters. After 25 iterations, the temperature drops to 100. At this time, the mine monitoring objective function takes a value of 12.5 to reach the minimum (the detection coverage A is 0.9, the monitoring cost Cost is 1.3 million yuan, and the adjustment range Δ adjust4.5), output the optimal network parameters, and set them according to the optimal network parameters to obtain the optimized perception network;

[0109] The time threshold is set to 1 month, and the engineering difference threshold is set to 10%.

[0110] In this embodiment, the method for performing a type of mine external early warning includes:

[0111] Optimize the external perception network through low-power wide area network connection to transmit external perception data to the edge end, and optimize the internal perception network through 5G private network connection to transmit internal perception data to the cloud platform;

[0112] Performing edge computing on external perception data at the edge to obtain edge processing results, specifically including: inputting the external perception data into the mine safety graph neural network to obtain external impact results, and inputting the external perception data into the mine safety prediction model to obtain internal and external correlation results; the edge processing results include external impact results and internal and external correlation results; the external impact results include risk categories and risk probabilities; the internal and external correlation results include predicted values for roof pressure distribution and groundwater level predictions;

[0113] Based on the external impact results, a type of mine external warning is performed at the edge end; the type of mine external warning includes: when the risk probability is greater than the first probability threshold, an external low-risk warning is performed; when the risk probability is greater than the second probability threshold, an external medium-risk warning is performed; when the risk probability is greater than the third probability threshold, an external high-risk warning is performed;

[0114] In the actual assessment, a risk warning was conducted for a certain mine. External perception data was input into the mine safety graph neural network to obtain external impact results. Taking the A15 grid as an example, the ventilation risk probability was 0.5, the equipment operation risk was 0.3, the mine tunnel stress change risk was 0.71, and the groundwater level rise risk was 0.59. The first probability threshold, the second probability threshold, and the third probability threshold were 0.4, 0.55, and 0.7, respectively. Therefore, a low-risk external warning due to ventilation, a medium-risk external warning due to groundwater level rise, and a high-risk external warning due to mine tunnel stress change were issued at the edge of the A15 grid.

[0115] External perception data is input into the mine safety prediction model to obtain internal and external correlation results. Taking the A15 grid as an example, the predicted value of roof pressure distribution is 12MPa and the predicted value of groundwater level is 105m.

[0116] In this embodiment, the method for obtaining and correcting internal perception data includes:

[0117] The edge processing results are transmitted to the cloud platform via the 5G private network. The roof pressure distribution prediction value of the internal and external correlation results is fused with the roof pressure of the internal sensing data to obtain the corrected roof pressure. The groundwater level prediction value of the internal and external correlation results is fused with the groundwater level of the internal sensing data to obtain the corrected groundwater level.

[0118] Extract the ventilation risk probability from the external impact results, calculate the deviation of gas concentration influencing factors based on external and internal sensing data, determine the ventilation factor based on the deviation of gas concentration influencing factors, and determine the corrected gas concentration based on the ventilation factor, ventilation risk probability, and gas concentration. The deviation of gas concentration influencing factors includes internal and external temperature deviation, internal and external humidity deviation, internal and external air pressure deviation, and internal and external wind speed deviation. The ventilation factor expression is:

[0119]

[0120] Among them F v is the ventilation factor, δ1 is the temperature and humidity weight, δ2 is the airflow weight, ΔT is the internal and external temperature deviation, ΔH is the internal and external humidity deviation, ΔP is the internal and external pressure deviation, P0 is the pressure reference value, ΔV is the internal and external wind speed deviation, V0 is the wind speed reference value;

[0121] Extract the equipment risk probability from the external impact results, divide the equipment data of the internal perception data into equipment operation data and equipment status data, determine the equipment status factor based on the equipment status data, and determine the modified equipment operation data based on the equipment status factor, equipment risk probability and equipment operation data; the equipment status factor expression is:

[0122]

[0123] Among them F m is the device status factor, Q is the number of devices, μ l is the maintenance weight of equipment l, is the location weight of equipment l in the mine tunnel, N l is the number of maintenance times of equipment l in one year, N l,0 is the industry benchmark annual maintenance frequency of equipment l, t l The duration of the most recent single maintenance of equipment l. is the average repair time of device l, τ l is the material degradation coefficient of equipment l, t l,tol is the cumulative running time of device l, t l,con is the continuous operation time of device l, t l,des is the designed operating time of device 1;

[0124] Performing image processing on the mine tunnel image to obtain mine tunnel image features; the mine tunnel image features include the number of mine tunnel cracks, the number and width of mine tunnel cracks, roof deformation value, equipment deformation value and foreign matter conditions;

[0125] The corrected roof pressure, corrected groundwater level, corrected gas concentration, corrected equipment operation data and mine tunnel image features are combined to form the corrected internal perception data;

[0126] In an actual assessment, a risk warning was issued for a certain mine. Taking the internal sensor data within the A15 grid as an example, the edge processing results were transmitted to the cloud platform via the 5G private network. The roof pressure distribution prediction value of 12MPa from the internal and external correlation results was fused with the roof pressure of 11MPa from the internal sensing data to obtain a corrected roof pressure of 11.5MPa. The groundwater level prediction value of 100m was fused with the groundwater level of 102m from the internal sensing data to obtain a corrected groundwater level of 103.5m.

[0127] The internal and external temperature deviation is calculated to be 2°C, the internal and external humidity deviation is 5%, the internal and external pressure deviation is 1 kPa, and the internal and external wind speed deviation is 1 m / s. The temperature and humidity weight δ1 is 0.4, the airflow weight δ2 is 0.6, the pressure reference value P0 is 100 kPa, and the wind speed reference value V0 is 4 m / s. The ventilation factor is calculated to be 1.27. Based on the ventilation risk probability of 0.5 and the gas concentration of 0.6%, the corrected gas concentration is determined to be 0.6% * (1 + 0.1 * 0.5 * 1.27) = 0.638%;

[0128] The equipment status factor is calculated based on the equipment data from the internal sensing data. Taking the roadheader as an example, the corrected cutting motor power is determined to be 320*(1+0.1*0.3*1.35=332.96W) based on the corresponding equipment status factor of 1.35, the cutting motor power of 320W, and the equipment operation risk of 0.3.

[0129] The mine tunnel image in the A15 grid was processed to obtain the mine tunnel image features, which showed that there were 3 cracks and the crack width was 2 mm.

[0130] In this embodiment, the method for performing internal early warning of a Class II mine includes:

[0131] Dynamic multi-sphere screening is used to correct internal perception data. The corrected internal perception data outside the multi-sphere is defined as internal risk warning data. The ratio of different types of internal risk warning data to the corresponding sphere radius is calculated to obtain different types of internal risk degrees. The radius update expression of the multi-sphere is:

[0132]

[0133] Where r(t) is the updated sphere radius at time t, r0 is the initial sphere radius, θ tis the time attenuation factor, λ2 is the density weight, λ3 is the abnormal data weight, ρ(t) is the data density of the sphere coverage area at the current moment, ∈ is the smoothing coefficient, N ano (t) is the count of abnormal data in the current sphere;

[0134] Two types of internal mine warnings are issued in the cloud based on different internal risk levels; the two types of internal mine warnings include: internal low-risk warnings when the risk level is greater than a first risk threshold, internal medium-risk warnings when the risk level is greater than a second risk threshold, and internal high-risk warnings when the risk level is greater than a third risk threshold;

[0135] In the actual assessment, a risk warning is carried out for a certain mine. Taking the corrected internal perception data (roof pressure) in the A15 grid as an example, the initial sphere radius is 8, the time attenuation factor is 0.5, the density weight is 0.3, and the abnormal data weight is 0.2. The data density of the area covered by the sphere at the current moment is 0.8, the smoothing coefficient is 0.1, the abnormal data count in the current sphere is 2, the update time is 5 days, and the update sphere radius is 11.04. The corresponding internal risk of the roof pressure corresponding to the corrected roof pressure of 11.5MPa is 1.042. According to the risk thresholds (first risk threshold 0.8, second risk threshold 0.95, third risk threshold 1.1), the internal risk warning of the roof pressure is carried out. Similarly, the internal risk of the gas concentration is calculated to be 0.83, the internal risk of the groundwater level is 0.92, and the internal risk of the mine tunnel structure is 1.12.

[0136] Therefore, the cloud platform issues internal low-risk warnings caused by gas concentration, internal low-risk warnings caused by groundwater level, internal medium-risk warnings caused by roof pressure, and internal high-risk warnings caused by mine tunnel structure to the corresponding monitoring nodes in the A15 grid.

[0137] In this embodiment, the method for performing three types of mine safety early warning includes:

[0138] The first-class mine external warning results are transmitted to the cloud platform via the 5G private network. The grid area where the first-class mine external warning results are located is matched with the sensor location corresponding to the second-class mine internal warning results. Three types of mine safety warnings are performed on the cloud for the same grid area. The first-class mine external warning and the second-class mine internal warning do not affect each other. The three types of mine safety warnings include mine safety low-risk warning, mine safety medium-risk warning, mine safety high-risk warning and mine safety emergency risk warning.

[0139] The strategies for the three types of mine safety warnings are as follows: when the risk levels of the first type of mine external warning and the second type of mine internal warning are the same, the mine safety warning risk level is increased; when the risk levels of the first type of mine external warning and the second type of mine internal warning are different, the higher risk level of the two is selected as the mine safety warning risk level; when the risk level of any type of warning is high risk and the risk level of the other type of warning is low risk or above, the mine safety warning risk level is defined as an emergency risk;

[0140] In the actual assessment, taking the risk warning of the A15 grid of a certain mine as an example, a medium-risk warning for mine safety caused by gas ventilation is given based on the external low-risk warning caused by ventilation and the internal low-risk warning caused by gas concentration. A medium-risk warning for mine safety caused by groundwater level is given based on the external medium-risk warning caused by rising groundwater level and the internal low-risk warning caused by groundwater level. An emergency risk warning for mine safety caused by stress change in mine tunnel is given based on the external high-risk warning caused by stress change in mine tunnel, the internal medium-risk warning caused by roof pressure and the internal high-risk warning caused by mine tunnel structure.

[0141] Secondly, a mine safety risk early warning system based on the Internet of Things includes:

[0142] Network parameter module: used to determine the mine monitoring objective function, set the initial perception network and optimize the network parameters according to the mine monitoring objective function to obtain an optimized perception network, and obtain internal perception data and external perception data according to the optimized perception network;

[0143] Edge computing module: used to build a mine safety graph neural network and a mine safety prediction model based on historical environmental safety data, input the external perception data into the mine safety graph neural network and the mine safety prediction model to obtain external impact results and internal and external correlation results, and perform a type of mine external early warning based on the external impact results;

[0144] Cloud computing module: used to correct the internal perception data according to the external influence result and the internal and external correlation result to obtain corrected internal perception data, use dynamic multi-sphere to filter the corrected internal perception data, and perform Class II mine internal warning according to the filtering result, and perform Class III mine safety warning by matching the results of Class I mine external warning and Class II mine internal warning;

[0145] Management module: used to manage the results of the first-class mine external warning, the second-class mine internal warning and the third-class mine safety warning, and perform mine safety management according to the warning results.

[0146] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A mine safety risk early warning method based on the Internet of Things, characterized in that: The following steps are involved: S1. Acquire historical environmental safety data, and construct a mine safety graph neural network and a mine safety prediction model based on the historical environmental safety data; S2. Determine a mine monitoring objective function, set up an initial perception network, and optimize network parameters according to the mine monitoring objective function to obtain an optimized perception network; the perception network includes an internal perception network and an external perception network; the network parameters include sensor locations and sensor quantity; S3. Acquire internal perception data and external perception data according to the optimized perception network, and input the external perception data into the mine safety graph neural network and the mine safety prediction model to obtain external impact results and internal and external correlation results; the internal perception data is transmitted via a 5G private network; and the external perception data is transmitted via a wide area network; S4. Performing a Class I mine external warning based on the external impact result, correcting the internal perception data based on the external impact result and the internal-external correlation result to obtain corrected internal perception data, screening the corrected internal perception data using a dynamic multi-sphere, and performing a Class II mine internal warning based on the screening result; S5. Conduct Class III mine safety warnings based on the external warning results of Class I mines and the internal warning results of Class II mines.

2. The mine safety risk early warning method based on the Internet of Things according to claim 1 is characterized in that: The method for constructing a mine safety graph neural network and a mine safety prediction model includes: Acquire historical environmental safety data via a low-power wide area network, including mine structural parameters, meteorological data, geological data, activity data, historical ventilation and drainage and equipment anomaly records, historical rock formation damage records, and historical support structure deformation records; Perform spatiotemporal alignment of historical environmental safety data, linking meteorological data with historical ventilation, drainage, and equipment anomaly records, geological data with historical rock formation damage records, and activity data with historical support structure deformation records; Constructing a mine safety graph neural network, which includes a dynamic graph structure, a graph attention network, and a spatiotemporal graph convolution module; The nodes of the dynamic graph structure include spatial region nodes and data entity nodes, and the edge relationships include spatial affiliation edges, causal influence edges, and risk transmission edges. The graph attention network is used to aggregate node features and perform cross-modal feature interaction on meteorological, geological, and activity data. The spatiotemporal graph convolution module extracts dynamic risk propagation patterns and outputs the risk category and probability of each node. Cross entropy loss and KL divergence are used as loss functions to constrain the consistency between the predicted risk distribution and historical disaster records, and the associated historical environmental safety data are used to train and evaluate the mine safety graph neural network. Constructing a mine safety prediction model, wherein the mine safety prediction model includes a roof pressure prediction base model and a groundwater level prediction base model; The roof pressure prediction base model receives mine structural parameters, geological data, activity data and corresponding associated data, and uses a dual-channel LSTM-BP hybrid structure to extract temporal features and static features to predict the distribution of roof pressure inside the mine; The groundwater level prediction base model receives mine structural parameters, meteorological data and corresponding associated data, uses the Wavelet-BP network to separate the trend term and periodic term of water level change through wavelet decomposition, and predicts the water level increase and water inrush risk level at different mining depths.

3. The mine safety risk early warning method based on the Internet of Things according to claim 1 is characterized in that: The method for obtaining an optimized perception network includes: The internal sensing network places multi-source sensors inside the mine to acquire internal sensing data; the internal sensing data includes gas concentration, roof pressure, groundwater level, equipment data, and mine tunnel images; the external sensing network places multi-source sensors outside the mine to acquire external sensing data; the external sensing data includes meteorological data, geological data, and activity data; The objective function of mine monitoring is determined by maximizing the monitoring range, minimizing the layout and operation costs, and minimizing the adjustment range. The expression is: Where Aim is the mine monitoring objective function, α is the monitoring coverage weight, A is the detection coverage, β is the cost weight, Cost is the monitoring cost, λ1 is the adjustment penalty coefficient, Δ adjust For adjustment range, M=M in +M out is the number of sensor categories in the perception network, which is equal to the number of sensor categories in the internal perception network M in and external sensing network sensor category M out The sum, w r is the perception network weight, R = {in, out} is the perception network set, w in is the internal perception network weight, w out is the external perception network weight, w m is the weight of m type sensor, N m is the number of m type sensors, A m is the monitoring range of type m sensor, A m,sta is the rated monitoring range of m-type monitoring data, Overlap(·) is the penalty term for the overlapping area of the internal and external networks, and A in is the coverage area of the internal sensor network, A out is the coverage area of the external sensor network, t i 、 are respectively the deployment cost, energy cost per unit time, running time and number of type i sensors in the internal perception network, t j 、 are respectively the deployment cost, energy cost per unit time, running time and number of type i sensors in the external perception network, is the Euclidean distance of the position change of m types of sensors, To adjust the position of the m-type sensor, To adjust the position of the front m type sensor, is the number of m type sensors after adjustment, To adjust the number of sensors in the first m categories; Set up the initial perception network, use the network parameters of the initial perception network as the initial population, and use the particle swarm-simulated annealing algorithm to optimize the network parameters to obtain the optimal network parameters. The specific steps are as follows: Initialize the population and update the particle speed and position. The expression is: in is the velocity update of particle i in dimension d at the k+1 iteration, is the speed corresponding to k iterations, is the position update of particle i in dimension d at k+1 iterations, is the position corresponding to k iterations, ω0 is the inertia weight, T k+1 、T k are the annealing temperatures for k+1 iterations and k iterations respectively, c1 and c2 are learning factors, r1 and r2 are random numbers in the range of (0,1), is the individual extreme value of particle i in dimension d in k iterations, that is, the individual optimal position of k iterations, is the global extreme value of the population in dimension d in k iterations, that is, the optimal position of the population in k iterations, ξ k+1 ,ξ k are the adaptive perturbation intensities for k+1 iterations and k iterations respectively, is a Gaussian random disturbance with variance σ 2 =T k , T0 is the initial temperature, v t is the basic cooling rate, is the particle swarm position variance measure for k iterations, is the number of particles in the particle swarm, which is equal to the number of sensors, P accept is the acceptance probability, Aim(X new )、Aim(X old ) are the objective functions of the new population and the old population respectively; The dual Metropolis criterion is used to update the individual extreme value and the global extreme value simultaneously. The expression is: in are the individual extreme value and global extreme value updated by k+1 iterations respectively, is the objective function of the position corresponding to the k+1 iteration of individual i in dimension d, are the objective functions of the individual extreme values of individual i in dimension d at k iterations and k+1 iterations, respectively. is the objective function value of the global extreme value of the population in dimension d at k iterations; Determine the network parameters based on the updated population position and calculate the mine monitoring objective function. Repeat the iteration until the mine monitoring objective function is minimized and output the optimal network parameters. Set them according to the optimal network parameters to obtain the optimized perception network. A time threshold and a project difference threshold are set. When any of the thresholds is met, the network optimization operation is triggered and the network parameters are updated. The project difference threshold represents the difference between the current mine project progress and the mine project progress at the last network parameter update time.

4. The mine safety risk early warning method based on the Internet of Things according to claim 1 is characterized in that: The method for performing a type of mine external early warning comprises: Optimize the external perception network through low-power wide area network connection to transmit external perception data to the edge end, and optimize the internal perception network through 5G private network connection to transmit internal perception data to the cloud platform; Performing edge computing on external perception data at the edge to obtain edge processing results, specifically including: inputting the external perception data into the mine safety graph neural network to obtain external impact results, and inputting the external perception data into the mine safety prediction model to obtain internal and external correlation results; the edge processing results include external impact results and internal and external correlation results; the external impact results include risk categories and risk probabilities; the internal and external correlation results include predicted values for roof pressure distribution and groundwater level predictions; A type of mine external warning is performed at the edge end based on the external impact results; the type of mine external warning includes: an external low-risk warning when the risk probability is greater than the first probability threshold, an external medium-risk warning when the risk probability is greater than the second probability threshold, and an external high-risk warning when the risk probability is greater than the third probability threshold.

5. The mine safety risk early warning method based on the Internet of Things according to claim 1 is characterized in that: The method for obtaining and correcting internal perception data comprises: The edge processing results are transmitted to the cloud platform via the 5G private network. The roof pressure distribution prediction value of the internal and external correlation results is fused with the roof pressure of the internal sensing data to obtain the corrected roof pressure. The groundwater level prediction value of the internal and external correlation results is fused with the groundwater level of the internal sensing data to obtain the corrected groundwater level. Extract the ventilation risk probability from the external impact results, calculate the deviation of gas concentration influencing factors based on external and internal sensing data, determine the ventilation factor based on the deviation of gas concentration influencing factors, and determine the corrected gas concentration based on the ventilation factor, ventilation risk probability, and gas concentration. The deviation of gas concentration influencing factors includes internal and external temperature deviation, internal and external humidity deviation, internal and external air pressure deviation, and internal and external wind speed deviation. The ventilation factor expression is: Among them F v is the ventilation factor, δ1 is the temperature and humidity weight, δ2 is the airflow weight, ΔT is the internal and external temperature deviation, ΔH is the internal and external humidity deviation, ΔP is the internal and external pressure deviation, P0 is the pressure reference value, ΔV is the internal and external wind speed deviation, V0 is the wind speed reference value; Extract the equipment risk probability from the external impact results, divide the equipment data of the internal perception data into equipment operation data and equipment status data, determine the equipment status factor based on the equipment status data, and determine the modified equipment operation data based on the equipment status factor, equipment risk probability and equipment operation data; the equipment status factor expression is: Among them F m is the device state factor, Q is the number of devices, μ l is the maintenance weight of equipment l, is the location weight of equipment l in the mine tunnel, N l is the number of maintenance times of equipment l in one year, N l,0 is the industry benchmark annual maintenance frequency of equipment l, t l The duration of the most recent single maintenance of equipment l. is the average repair time of device l, τ l is the material degradation coefficient of equipment l, t l,tol is the cumulative running time of device l, t l,con is the continuous operation time of device l, t l,des is the designed operating time of device 1; Performing image processing on the mine tunnel image to obtain mine tunnel image features; the mine tunnel image features include the number of mine tunnel cracks, the number and width of mine tunnel cracks, roof deformation value, equipment deformation value and foreign matter conditions; The corrected roof pressure, corrected groundwater level, corrected gas concentration, corrected equipment operation data and mine tunnel image features are combined to form the corrected internal perception data.

6. The mine safety risk early warning method based on the Internet of Things according to claim 1 is characterized in that: The method for conducting internal early warning of a Class II mine includes: Dynamic multi-sphere screening is used to correct internal perception data. The corrected internal perception data outside the multi-sphere is defined as internal risk warning data. The ratio of different types of internal risk warning data to the corresponding sphere radius is calculated to obtain different types of internal risk degrees. The radius update expression of the multi-sphere is: Where r(t) is the updated sphere radius at time t, r0 is the initial sphere radius, θ t is the time attenuation factor, λ2 is the density weight, λ3 is the abnormal data weight, ρ(t) is the data density of the sphere coverage area at the current moment, ∈ is the smoothing coefficient, N ano (t) is the count of abnormal data in the current sphere; Two types of internal mine warnings are carried out in the cloud based on different types of internal risk levels; the two types of internal mine warnings include: internal low-risk warnings when the risk level is greater than the first risk level threshold, internal medium-risk warnings when the risk level is greater than the second risk level threshold, and internal high-risk warnings when the risk level is greater than the third risk level threshold.

7. The mine safety risk early warning method based on the Internet of Things according to claim 1 is characterized in that: The method for performing three types of mine safety early warning includes: The first-class mine external warning results are transmitted to the cloud platform via the 5G private network. The grid area where the first-class mine external warning results are located is matched with the sensor location corresponding to the second-class mine internal warning results. Three types of mine safety warnings are performed on the cloud for the same grid area. The first-class mine external warning and the second-class mine internal warning do not affect each other. The three types of mine safety warnings include mine safety low-risk warning, mine safety medium-risk warning, mine safety high-risk warning and mine safety emergency risk warning. The strategies for the three types of mine safety warnings are as follows: when the risk levels of the external warning of a type one mine and the internal warning of a type two mine are the same, the risk level of the mine safety warning is increased; when the risk levels of the external warning of a type one mine and the internal warning of a type two mine are different, the higher risk level of the two is selected as the mine safety warning risk level; when the risk level of any type of warning is high risk and the risk level of the other type of warning is low risk or above, the mine safety warning risk level is defined as an emergency risk.

8. A mine safety risk early warning system based on the Internet of Things, used to execute the method according to any one of claims 1 to 7, characterized in that: include: Network parameter module: used to determine the mine monitoring objective function, set the initial perception network and optimize the network parameters according to the mine monitoring objective function to obtain an optimized perception network, and obtain internal perception data and external perception data according to the optimized perception network; Edge computing module: used to build a mine safety graph neural network and a mine safety prediction model based on historical environmental safety data, input the external perception data into the mine safety graph neural network and the mine safety prediction model to obtain external impact results and internal and external correlation results, and perform a type of mine external early warning based on the external impact results; Cloud computing module: used to correct the internal perception data according to the external influence result and the internal and external correlation result to obtain corrected internal perception data, use dynamic multi-sphere to filter the corrected internal perception data, and perform Class II mine internal warning according to the filtering result, and perform Class III mine safety warning by matching the results of Class I mine external warning and Class II mine internal warning; Management module: used to manage the results of the first-class mine external warning, the second-class mine internal warning and the third-class mine safety warning, and perform mine safety management according to the warning results.

Images