An artificial intelligence-based mine disaster period ventilation emergency decision system

Abstract

The application discloses a mine disaster period ventilation emergency decision system based on artificial intelligence, which comprises a data acquisition module, a disaster perception module, a decision optimization module and a decision execution module.The data acquisition module acquires multi-source real-time data in the mine disaster period, and obtains standardized mine space-time data through data cleaning and alignment preprocessing operations.The disaster perception module extracts spatial and data correlation information to construct a mine topology graph structure based on the standardized space-time data set, and constructs a dynamic three-dimensional air flow field through a graph neural network based on the mine topology graph structure.The decision optimization module regards the dynamic three-dimensional air flow field as a dynamic changing physical background field, defines a disaster event as a disturbance source in the physical background field, and deduces disaster evolution based on a core disturbance propagation rule to obtain disaster situation awareness data.The decision execution module performs mine disaster early warning based on the disaster situation awareness data, and realizes the best ventilation emergency decision in the mine disaster period.

Description

Technical Field

[0001] This invention relates to the field of next-generation information technology, and in particular to an artificial intelligence-based emergency decision-making system for ventilation during mine disasters. Background Technology

[0002] The evolution of mine ventilation emergency decision-making technology has always been related to mine safety needs. In the early stages, due to the lack of monitoring methods, ventilation emergency decision-making mainly relied on human experience and static plans. Technicians formulated fixed emergency plans based on basic mine design parameters such as roadway direction and fan power. When faced with sudden disasters such as gas leaks and fires, they could only adjust ventilation equipment according to the preset procedures and could not adapt to the dynamic evolution characteristics of disasters.

[0003] Existing ventilation emergency decision-making systems mostly use pure data-driven neural network models, relying solely on monitoring data for feature fitting without incorporating the core physical laws of mine ventilation, such as the conservation of air volume and the relationship between air pressure difference and resistance loss. This modeling method is prone to physical distortion problems, such as the deduced airflow direction contradicting the actual roadway ventilation law, or the air pressure distribution not conforming to the principles of fluid mechanics, resulting in a large deviation between the disaster evolution deduction results and the actual situation.

[0004] Furthermore, existing extrapolation methods often treat disaster parameters as independently diffused substances, neglecting the forced carrying effect of the dynamic airflow field in the mine on disaster diffusion. In reality, gas is quickly carried to the return airway by the airflow, but traditional models often predict its uniform diffusion, which cannot accurately depict the dual evolution law of natural diffusion and airflow drive. This results in insufficient accuracy in predicting the scope and diffusion rate of disaster impact, making it difficult to support accurate early warning.

[0005] Therefore, this paper proposes an artificial intelligence-based emergency decision-making system for ventilation during mine disasters. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above objectives, the present invention proposes the following technical solution:

[0007] An artificial intelligence-based emergency decision-making system for ventilation during mine disasters includes:

[0008] Data acquisition module: Collects multi-source real-time data during mine disasters, and obtains standardized mine spatiotemporal data through data cleaning and alignment preprocessing operations;

[0009] Disaster perception module: Based on standardized spatiotemporal datasets, it extracts spatial and data correlation information to construct a mine topology map structure. Based on the mine topology map structure, it constructs a dynamic three-dimensional airflow field through a graph neural network that integrates physical constraints.

[0010] The graph neural network that integrates physical constraints is achieved by introducing a first physical constraint and a second physical constraint, embedding the first physical constraint and the second physical constraint into the message passing process of the original graph neural network;

[0011] Decision optimization module: Treats the dynamic three-dimensional airflow field as a dynamically changing physical background field, defines the disaster event as a disturbance source in the physical background field, and deduces the disaster evolution based on the core disturbance propagation rules to obtain disaster situation awareness data.

[0012] Decision execution module: Based on disaster situation awareness data, it conducts mine disaster early warning and realizes the best ventilation emergency decision during the mine disaster period.

[0013] The process of obtaining standardized mine spatiotemporal data is as follows:

[0014] Deploy a sensor network to collect wind speed data and gas concentration data, and perform time alignment. Use timestamp t as a unified benchmark to match the time dimension of wind speed data and gas concentration data.

[0015] Compare the acquisition time of each data point with the system's standard time. If the deviation exceeds [a certain threshold], [the system will take appropriate action]. If so, adjust its corresponding timestamp to the nearest standard timestamp to form time-aligned data;

[0016] Based on the spatial coordinates (x, y, z) of the monitoring points, time-aligned data is associated with spatial location. For each monitoring point, its corresponding wind speed data and gas concentration data are bound to the spatial coordinates (x, y, z) to form a three-dimensional data structure of spatial coordinates, time, and parameters, thereby obtaining standardized mine spatiotemporal data.

[0017] The process of constructing the mine topology map structure is as follows:

[0018] Based on a standardized spatiotemporal dataset, the spatial coordinates of all monitoring points and ventilation equipment are extracted. The spatial coordinates of the ventilation equipment are associated with their corresponding equipment numbers. Each monitoring point and ventilation equipment is treated as an independent node, and a graph node set is constructed. Based on the actual roadway connections and spatial distances in the mine, a graph edge set E is constructed, and a graph node set E is used. With edge set Construct a mine topology map .

[0019] The first physical constraint describes the conservation of air volume in any spatial region per unit time, for a mine topology map. any node in Within the corresponding spatial area, the total inflow air volume equals the total outflow air volume, as shown in the formula: ;

[0020] in, For nodes The set of adjacent nodes, For nodes and The air volume between.

[0021] The second physical constraint describes the relationship between air pressure difference, wind speed, and resistance loss within the roadway, for a mine topology map. any edge in The corresponding tunnel has a wind pressure difference that satisfies the formula:

[0022] ;

[0023] in, For nodes Standardized wind pressure, For nodes Standardized wind pressure, For nodes wind speed, For nodes wind speed, This refers to the air density inside the mine. This is the roadway roughness coefficient, pre-calibrated based on the wear degree of the roadway wall material. The equivalent diameter of the tunnel. The average wind speed in the tunnel. This refers to the length of the tunnel.

[0024] The process of constructing a dynamic three-dimensional airflow field is as follows:

[0025] The first and second physical constraints are embedded into the message passing process of the graph neural network. Let the total number of layers in the graph neural network be . ;

[0026] Among them, the node features of the initial layer are the initial features of the nodes in the mine topology graph G;

[0027] Then, through iterative updates using a graph neural network, until the... Once all node features of the layer converge, the corresponding wind speed and wind pressure data are extracted to obtain the field wind speed and field wind pressure values.

[0028] The wind speed and wind pressure values ​​at all spatial locations of all nodes are obtained, and spatial interpolation methods are used to complete the wind speed and wind pressure values ​​at all spatial locations within the mine, forming a dynamic three-dimensional airflow field.

[0029] The process of constructing the core perturbation propagation rules is as follows:

[0030] Disturbance propagation is decomposed into two core physical processes: natural diffusion and wind-borne propagation. The core disturbance propagation rule is expressed as follows: ;

[0031] in, This represents the core catastrophic parameter, where D is the diffusion coefficient. Indicates the process of airflow carrying, Indicates the natural diffusion process. Indicates the source of catastrophic disturbance. The intensity of the disturbance source.

[0032] The process of obtaining disaster situation awareness data is as follows:

[0033] Dynamic three-dimensional airflow field The physical background field of the mine environment is regarded as the physical background field, and the disaster event is defined as the disaster disturbance source in the physical background field. The evolution is deduced according to the core disturbance propagation rule, and the core disaster parameters of the mine disaster are recorded at each grid point (x,y,z,t) at each time stamp t.

[0034] The core disaster parameters include disaster gas concentration, field wind speed deviation, and field wind pressure deviation.

[0035] By setting deviation thresholds to label risk levels, core disaster parameters for each region are integrated with their corresponding risk levels to form disaster situation awareness data. .

[0036] The present invention has the following beneficial effects:

[0037] 1. By combining the core physical laws of mine ventilation with graph neural networks, two physical constraints are introduced and embedded into the message passing process of the graph neural network. On the one hand, the node feature update logic is corrected by ensuring that the total inflow of air to any node equals the total outflow of air, so as to ensure that the reconstructed airflow field conforms to the principle of mass conservation. On the other hand, the physical constraint loss is calculated by the relationship between the roadway air pressure difference and wind speed and roughness coefficient, and the feature parameters that do not conform to the laws of fluid mechanics are penalized. Through the iterative update of the multi-layer graph neural network until the node features converge, and then combined with spatial interpolation to complete the wind speed and wind pressure data of the entire mine, the final dynamic three-dimensional airflow field has both data fitting accuracy and physical rationality, accurately presenting the airflow differences of different roadways, and providing a physical background field that is highly consistent with the actual mine environment for subsequent disaster evolution simulation, avoiding the simulation deviation caused by the traditional model deviating from the physical laws.

[0038] 2. By considering the dynamic three-dimensional airflow field as the physical background field for disaster evolution, and defining disasters such as gas leaks and fires as disturbance sources, a core disturbance propagation rule integrating natural diffusion and airflow carrying is designed. The diffusion term characterizes the natural diffusion trend of parameters such as gas, while the airflow carrying term reflects the driving effect of dynamic airflow on disaster diffusion, accurately capturing the spatiotemporal evolution law of disaster parameters. On this basis, with gas concentration, wind speed deviation, and wind pressure deviation as core indicators, a multi-level risk threshold is set to form a hierarchical early warning mechanism. When medium-risk parameters appear in a local area, a level-three early warning is triggered and the personnel evacuation procedure is initiated. This not only identifies disaster threats at an early stage, but also presents the secondary risks of ventilation system disturbances through multi-parameter linkage, solving the problems of incomplete and delayed response of traditional single-parameter early warning. Attached Figure Description

[0039] Figure 1 This is a system block diagram of an artificial intelligence-based emergency decision-making system for ventilation during mine disasters, as proposed in this invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] Example: Figure 1 As shown, the present invention proposes an artificial intelligence-based emergency decision-making system for ventilation during mine disasters, comprising:

[0042] Data acquisition module: Collects multi-source real-time data during mine disasters, and obtains standardized mine spatiotemporal data through data cleaning and alignment preprocessing operations;

[0043] Deploy a sensor network, using a distributed sensor network composed of wind speed sensors and gas concentration sensors already deployed in the mine, to collect wind speed data and gas concentration data in real time. All data are sampled synchronously according to time resolution to form the original data set D.

[0044] Perform targeted cleaning on the original dataset D to remove outliers and missing values.

[0045] For wind speed data, a reasonable range of values ​​is determined based on the physical operating limits of the mine ventilation system. If a data point exceeds this range, it is judged as an outlier. Outlier handling adopts the method of replacing the average of adjacent time points. For outlier data points, the average of the data at the previous time point and the data at the next time point is calculated and replaced.

[0046] For gas concentration data, a reasonable range of values ​​is determined based on the safety threshold of toxic and harmful gases. If a data point exceeds this range, it is judged as an outlier. Outlier handling adopts the mean replacement method of adjacent time points. For outlier data points, the mean of the data at the previous time point and the data at the next time point is calculated and replaced.

[0047] Then, data alignment is performed, using timestamp t as a unified benchmark, to match the wind speed data and gas concentration data in the time dimension.

[0048] Since all data are processed at a fixed time resolution While there is no time deviation in data acquisition, a slight delay may occur during actual transmission. Therefore, precise alignment is achieved through timestamp calibration.

[0049] Compare the acquisition time of each data point with the system's standard time. If the deviation exceeds [a certain threshold], [the system will take appropriate action]. Then, adjust the corresponding timestamp to the nearest standard timestamp to ensure that the wind speed and gas concentration data of all monitoring points are included under the same timestamp, thus forming time-aligned data. ;

[0050] Then, spatial alignment is performed, aligning the time data based on the spatial coordinates (x, y, z) of the monitoring points. By associating data with spatial location, for each monitoring point, its corresponding wind speed and gas concentration data are bound to spatial coordinates (x, y, z), forming a three-dimensional data structure of spatial coordinates, time, and parameters, i.e., standardized mine spatiotemporal data. ,in, For the standardized wind speed data at monitoring point i, This refers to the standardized wind pressure data for monitoring point i, where i represents the index of the monitoring point and the spatial coordinates corresponding to monitoring point i. .

[0051] Disaster perception module: Based on standardized spatiotemporal datasets, it extracts spatial and data correlation information to construct a mine topology map structure, and constructs a dynamic three-dimensional airflow field based on the mine topology map structure through a graph neural network that integrates physical constraints;

[0052] The process of constructing the mine topology map structure is as follows:

[0053] Firstly, based on a standardized spatiotemporal dataset, spatial and data correlation information is extracted to construct a mine topology map structure, from which standardized mine spatiotemporal data is derived. Extract the spatial coordinates of all monitoring points and ventilation equipment;

[0054] Specifically, the spatial coordinates of the monitoring points are directly taken from the spatial coordinates of all monitoring points.

[0055] Associating the spatial coordinates of ventilation equipment with their corresponding equipment numbers, and treating each monitoring point and ventilation equipment as an independent node, a graph node set is constructed. ,in, The node corresponding to detection point i;

[0056] Based on the actual roadway connections and spatial distances in the mine, construct a graph edge set E. For any two nodes... and , If the corresponding spatial locations of the two nodes are in the same or adjacent lanes, an edge is established between the two nodes, and the correlation strength of the edge is the spatial distance d between the two nodes, ultimately forming an edge set. ;

[0057] Through the graph node set With edge set Construct a mine topology map ;

[0058] Specifically, this topology graph Linking the spatial structure of the mine with monitoring and equipment data into a graph theory structure is the basis for subsequent message passing and feature updating using graph neural networks that integrate physical constraints.

[0059] The process of constructing a dynamic three-dimensional airflow field is as follows:

[0060] Based on mine topology map The physical constraints of mine airflow are embedded into the message passing process of a graph neural network, and the three-dimensional airflow field is dynamically reconstructed by iteratively updating node features.

[0061] First, determine the core physical constraints of mine ventilation, including the first physical constraint and the second physical constraint;

[0062] The first physical constraint describes the conservation of air volume in any spatial region per unit time, for a mine topology map. any node in Within the corresponding spatial area, the total inflow air volume equals the total outflow air volume, as shown in the formula:

[0063] ;

[0064] in, For nodes The set of adjacent nodes, For nodes and Wind speed between;

[0065] The second physical constraint describes the relationship between air pressure difference and wind speed and resistance loss within the roadway, for mine topology maps. any edge in The corresponding tunnel has a wind pressure difference that satisfies the formula:

[0066] ;

[0067] in, For nodes Standardized wind pressure, For nodes Standardized wind pressure, For nodes wind speed, For nodes wind speed, This refers to the air density inside the mine. This is the roadway roughness coefficient, pre-calibrated based on the wear degree of the roadway wall material. The equivalent diameter of the tunnel. The average wind speed in the tunnel. The length of the tunnel;

[0068] By embedding the first and second physical constraints into the message passing process of the graph neural network, a graph neural network integrating physical constraints is constructed. Let the total number of layers in the graph neural network be . The kth layer (k=1,2,…, The node feature update depends on the node features of the (k-1)th layer and the physical constraint loss;

[0069] The node characteristics of the initial layer (k=0) are the same as the initial characteristics of the nodes in the mine topology graph G;

[0070] For any node in the k-th layer, the update formula is:

[0071] ;

[0072] in, For node v at the th The feature vector of the layer, For the first The weight matrix of the layer, For ReLU function, As the first physical constraint, As a second physical constraint, For nodes The set of adjacent nodes;

[0073] Through iterative updates using a multi-layer graph neural network, until the... All node features of a layer converge, meaning the mean square error of the features of all adjacent nodes in the same layer is less than a preset threshold. At that time, extract the wind speed and wind pressure of all nodes, for each node The corresponding wind speed and wind pressure components are the field wind speed at that spatial location. With field wind pressure value Based on the field wind speed of all nodes With wind pressure value Spatial interpolation methods were used to complete the wind speed and wind pressure data for all spatial locations within the mine, forming a dynamic three-dimensional airflow field. ,in, Let be the field wind speed at any spatial coordinate (x, y, z) at time t. This represents the field wind pressure corresponding to the spatial coordinates.

[0074] Decision optimization module: Treats the dynamic three-dimensional airflow field as a dynamically changing physical background field, defines the disaster event as a disturbance source in the physical background field, and deduces the disaster evolution based on the core disturbance propagation rules to obtain disaster situation awareness data.

[0075] Dynamic three-dimensional airflow field Considered as the physical background field of the mine environment, it contains the core parameters of wind speed vector and wind pressure value at any spatial coordinate (x,y,z) at time t;

[0076] Specifically, because the three-dimensional airflow field is not static, the wind speed will change with the operating status of the ventilation equipment fan and damper, and the wind pressure will vary due to the roadway resistance and the start and stop of the equipment. Therefore, the dynamic three-dimensional airflow field is regarded as the physical background field and the basic environment for the evolution of disaster.

[0077] Then, catastrophic disturbance sources are defined, and catastrophic events are defined as catastrophic disturbance sources in the physical background field, such as gas leaks, fires, and roof collapses, denoted as... ,in, The spatial coordinates of the disaster occurrence are located by abnormal sensor data, such as the point of sudden increase in gas concentration. 't' is the timestamp of the disaster's onset, determined by the moment of abrupt change in the monitoring data. It refers to the intensity of the disturbance source, such as the leakage rate of gas leaks per unit time or the heat release per unit time of a fire.

[0078] Specifically, the core function of disturbance sources is to disrupt the stable state of the background field. For example, gas leaks will continuously inject gas into the background field, changing the local gas concentration, and roof collapses will block the roadway, forcing the wind speed and wind pressure in the background field to redistribute.

[0079] Construct a core perturbation propagation rule formula, with the physical background field as the basis. With the source of catastrophic disturbance Using the given core disturbance propagation rules as input, the parameter changes are calculated to deduce the catastrophe evolution process;

[0080] Specifically, traditional diffusion rules only consider the natural diffusion of materials, but the core contradiction in the mining scenario is the diffusion of catastrophic parameters (such as gas), which will be forcibly carried by the airflow of the ventilation system. If the role of airflow is ignored, the deduction results will deviate significantly from the actual catastrophic diffusion path. For example, in reality, gas will be quickly carried to the return airway by the airflow, but the traditional formula will predict its slow and uniform diffusion. Therefore, the design objective is clearly defined as the dual disturbance propagation law of diffusion plus airflow drive.

[0081] The disturbance propagation is decomposed into two core physical processes and corresponding design formula terms, including the natural diffusion process and the wind-borne process. The core disturbance propagation rule is that the disturbance source induces parameter changes, which follow the laws of diffusion and wind-borne processes in the background field. The core disturbance propagation rule is expressed as:

[0082] ;

[0083] in, This represents the core catastrophic parameter, where D is the diffusion coefficient. Indicates the process of airflow carrying, Indicates the natural diffusion process;

[0084] Specifically, core disaster parameters By physical background field The core parameters provided, such as wind speed vector u and wind pressure vector q, are substituted into its control equation. Through numerical calculation, the field distribution at each grid point in the entire three-dimensional space at each future time is gradually deduced.

[0085] Then, evolutionary deductions are performed based on the core perturbation propagation rules:

[0086] First, initial and boundary conditions are set; core disaster parameters are defined before the disaster occurs. Take the stable value of the background field, that is, set the physical starting point and constraints for numerical calculation;

[0087] The initial condition is the time when the disaster occurs. Compute the core disaster parameters for all locations within the domain. ;

[0088] The boundary conditions are set based on the physical characteristics of the mine roadway. The roadway wall is defined as a no-flux boundary, meaning the gradient of the parameter in the direction normal to the wall is 0. This indicates that matter or energy cannot pass through the roadway wall. ;

[0089] Then, numerical solutions were performed. The mine space was divided into a 1m×1m×1m three-dimensional grid, and time was discretized at 1-minute intervals. The finite difference method was used to iteratively calculate the time stamps of each grid point at different timestamps. Value, taking gas leakage as an example: from t= start, For the time of the disaster, each timestamp is substituted with the real-time wind speed of the background field, and the gas concentration at the grid points is calculated until a simulation duration of 60 minutes is covered.

[0090] Finally, the evolution results are recorded. The system records the core disaster parameters of the mine disaster at each grid point (x,y,z,t) at each time stamp t, including the disaster gas concentration. Field wind speed deviation Field wind pressure deviation This is used to quantify the degree of disturbance to the ventilation system;

[0091] Among them, field wind speed deviation Field wind pressure deviation Field wind speed at the same spatial location under different time conditions With wind pressure value The difference, specifically, is the difference between the value at the end of the disaster simulation and the initial normal value before the disaster occurred;

[0092] Then, based on prior knowledge, the core disaster parameters, such as gas concentration and field wind speed deviation, are set. Field wind pressure deviation The deviation threshold, for example:

[0093] Gas concentration: ≤0.8% (low risk), 0.8%-1.2% (medium risk), >1.2% (high risk);

[0094] Wind pressure deviation: ≤50Pa (low risk), 50-100Pa (medium risk), >100Pa (high risk);

[0095] The parameters for each region are labeled with risk levels and integrated into structured data. All risk levels and corresponding core disaster parameters are then combined to form disaster situation awareness data. ,in, The risk level is indicated by 1 for low risk, 2 for medium risk, and 3 for high risk.

[0096] Specifically, disaster situation awareness data In this system, wind speed and wind pressure deviations can provide early warnings of potential risks. Larger values ​​for these deviations indicate more severe disturbances to the system caused by disasters such as roadway blockages or fan malfunctions.

[0097] By correction and ,make and Returning to near zero can restore the stability of the ventilation system, thereby controlling the spread of gas concentration. and It is a direct indicator for assessing the effectiveness of emergency decisions;

[0098] Meanwhile, the catastrophic gas concentration is obtained based on grid points (x,y,z,t) through the core perturbation propagation rule. By statistically analyzing the grid areas where the catastrophic gas concentration exceeds the threshold at different time stamps, the spread path of the catastrophic event and the expansion rate of its affected area can be clearly presented. A change, such as rising from 0.9% to 1.1% in just 5 minutes, can reflect the speed of a disaster's development. The greater the change, the stronger the disturbance source, requiring more urgent intervention measures.

[0099] Decision execution module: Based on disaster situation awareness data, it conducts mine disaster early warning and realizes the best ventilation emergency decision-making during mine disasters;

[0100] Based on disaster situational awareness data, situational data for each grid point is extracted every 30 seconds, and thresholds are compared point by point.

[0101] When any area has ≥3 adjacent grid points with medium risk indicators, a Level 1 warning is triggered, a pop-up window will indicate that the risk in the area has increased, and the spatial range of the area will be displayed simultaneously, such as "section 150-170m of No. 3 return airway";

[0102] When a high-risk indicator appears in any area, a level-two warning is triggered, an audible and visual alarm is activated, and the personnel location card access in that area is automatically locked to prevent personnel from entering.

[0103] When two high-risk indicators appear simultaneously in any area, a Level 3 warning is triggered. The underground broadcast will continuously play evacuation instructions and push the disaster avoidance route of the area to all underground terminals.

[0104] The warning status is dynamically synchronized, and the disaster situation awareness data is updated every minute. If the indicators of the original high-risk area fall back to medium risk, the level two warning is downgraded to the level one warning.

[0105] If the indicators of the original medium-risk area remain stable below the warning value for 2 minutes, the Level 1 warning will be lifted and the pre-warning status will be restored.

[0106] If the indicators in the warning area continue to escalate, the warning level will be triggered directly.

[0107] In the application, several formulas are calculated by removing dimensions and taking their numerical values. The formulas are established by collecting a large amount of data and simulating the most recent real situation. Some coefficients or weights in the formulas are set by those skilled in the art according to the actual situation, so they will not be elaborated here.

[0108] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based emergency decision-making system for ventilation during mine disasters, characterized in that, include: Data acquisition module: Collects multi-source real-time data during mine disasters, and obtains standardized mine spatiotemporal data through data cleaning and alignment preprocessing operations; The process of obtaining standardized mine spatiotemporal data is as follows: Deploy a sensor network to collect wind speed data and gas concentration data, and perform time alignment. Use timestamp t as a unified benchmark to match the time dimension of wind speed data and gas concentration data. Compare the acquisition time of each data point with the system's standard time. If the deviation exceeds [a certain threshold], [the system will take appropriate action]. If so, adjust its corresponding timestamp to the nearest standard timestamp to form time-aligned data; Based on the spatial coordinates (x, y, z) of the monitoring points, the time-aligned data is associated with the spatial location. For each monitoring point, its corresponding wind speed data and gas concentration data are bound to the spatial coordinates (x, y, z) to form a three-dimensional data structure of spatial coordinates, time, and parameters, thereby obtaining standardized mine spatiotemporal data. Disaster perception module: Based on standardized spatiotemporal datasets, it extracts spatial and data correlation information to construct a mine topology map structure. Based on the mine topology map structure, it constructs a dynamic three-dimensional airflow field through a graph neural network that integrates physical constraints. The graph neural network that integrates physical constraints is achieved by introducing a first physical constraint and a second physical constraint, embedding the first physical constraint and the second physical constraint into the message passing process of the original graph neural network; The first physical constraint describes the conservation of air volume in any spatial region per unit time, for a mine topology map. any node in Within the corresponding spatial area, the total inflow air volume equals the total outflow air volume, as shown in the formula: ; in, For nodes The set of adjacent nodes, For nodes and Wind speed between; The second physical constraint describes the relationship between air pressure difference, wind speed, and resistance loss within the roadway, for a mine topology map. any edge in The corresponding tunnel has a wind pressure difference that satisfies the formula: ； in, For nodes Standardized wind pressure, For nodes Standardized wind pressure, For nodes wind speed, For nodes wind speed, This refers to the air density inside the mine. This is the roughness coefficient of the roadway. The equivalent diameter of the tunnel. The average wind speed in the tunnel. The length of the tunnel; Decision optimization module: Treats the dynamic three-dimensional airflow field as a dynamically changing physical background field, defines the disaster event as a disturbance source in the physical background field, and deduces the disaster evolution based on the core disturbance propagation rules to obtain disaster situation awareness data. Decision execution module: Based on disaster situation awareness data, it conducts mine disaster early warning and realizes the best ventilation emergency decision during the mine disaster period.

2. The mine disaster emergency ventilation decision-making system based on artificial intelligence as described in claim 1, characterized in that, The process of constructing the mine topology map structure is as follows: Based on a standardized spatiotemporal dataset, the spatial coordinates of all monitoring points and ventilation equipment are extracted. The spatial coordinates of the ventilation equipment are associated with their corresponding equipment numbers. Each monitoring point and ventilation equipment is treated as an independent node, and a graph node set is constructed. Based on the actual roadway connections and spatial distances in the mine, a graph edge set E is constructed, and a graph node set E is used. With edge set Construct a mine topology map .

3. The mine disaster emergency ventilation decision-making system based on artificial intelligence as described in claim 1, characterized in that, The process of constructing a dynamic three-dimensional airflow field is as follows: The first and second physical constraints are embedded into the message passing process of the graph neural network. Let the total number of layers in the graph neural network be . ; Among them, the node features of the initial layer are the initial features of the nodes in the mine topology graph G; Then, through iterative updates using a graph neural network, until the... Once all node features of the layer converge, the corresponding wind speed and wind pressure data are extracted to obtain the field wind speed and field wind pressure values. The wind speed and wind pressure values ​​at all spatial locations of all nodes are obtained, and spatial interpolation methods are used to complete the wind speed and wind pressure values ​​at all spatial locations within the mine, forming a dynamic three-dimensional airflow field.

4. The mine disaster emergency ventilation decision-making system based on artificial intelligence according to claim 3, characterized in that, The process of constructing the core perturbation propagation rules is as follows: Disturbance propagation is decomposed into two core physical processes: natural diffusion and wind-borne propagation. The core disturbance propagation rule is expressed as follows: ; in, This represents the core catastrophic parameter, where D is the diffusion coefficient. Indicates the process of airflow carrying, Indicates the natural diffusion process. Indicates the source of catastrophic disturbance. The intensity of the disturbance source.

5. The mine disaster emergency ventilation decision-making system based on artificial intelligence according to claim 4, characterized in that, The process of obtaining disaster situational awareness data is as follows: Dynamic three-dimensional airflow field The physical background field of the mine environment is regarded as the physical background field, and the disaster event is defined as the disaster disturbance source in the physical background field. The evolution is deduced according to the core disturbance propagation rule, and the core disaster parameters of the mine disaster are recorded at each grid point (x,y,z,t) at each time stamp t. The core disaster parameters include disaster gas concentration, field wind speed deviation, and field wind pressure deviation. By setting deviation thresholds to label risk levels, core disaster parameters for each region are integrated with their corresponding risk levels to form disaster situation awareness data. .

Images