Coal Mine Disaster Identification Method Based on Deep Learning Algorithm and Related Equipment

Abstract

The present application discloses a method for identifying coal mine disasters based on deep learning algorithms and related devices. The method includes: obtaining laser scanning point cloud data, geological drilling data, and geophysical exploration data of the coal mine; dividing the target area into multiple three-dimensional grid cells, and interpolating the three-dimensional grid cells to obtain a first three-dimensional geological model; adding rock layer and coal seam parameters to the first three-dimensional geological model to obtain a second three-dimensional geological model; associating the second three-dimensional geological model with a big data model to obtain monitoring data for the current season; performing feature extraction on the current season and the monitoring data to obtain a multi-dimensional time series; inputting the multi-dimensional time series into a deep learning model to obtain a water level prediction value; inputting the water level prediction value into the big data model for correction; spatially combining the corrected water level prediction value with the second three-dimensional geological model to obtain a dynamic three-dimensional geological model; and determining the probability of water disasters occurring in the coal mine based on the dynamic three-dimensional geological model.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of disaster identification, and in particular to a coal mine disaster identification method based on a deep learning algorithm and related devices. Background Art

[0002] The most common disaster in coal mines is water disaster at present. This is because the groundwater system is complex, and it is difficult to avoid the infiltration of surface water during the construction of coal mines. If accompanied by extreme weather events such as heavy rain and floods, it may cause potential water disaster hazards in the coal mine, which may damage the equipment and infrastructure in the coal mine. Moreover, it may trigger secondary disasters and cause casualties. Therefore, it is very necessary to monitor the potential water disaster hazards in the coal mine.

[0003] Currently, observation wells are mainly arranged around the coal mine, and the groundwater level of the observation wells is regularly measured by water level gauges installed in the observation wells, so as to obtain the change trend of the groundwater level. Then, the possibility of water disaster occurrence in the coal mine can be predicted through experience.

[0004] For the above monitoring method, the water level change of the observation well may lag behind the actual water situation change of the coal mine, because the groundwater level usually changes with seasons. For example, in the rainy season, the groundwater level may generally rise; in the dry season, the water level may drop. Therefore, when a sudden water situation occurs, the resulting water level change may be masked by seasonal changes. Therefore, seasonal groundwater level changes may mask the short-term changes caused by sudden water situations in the mine. This leads to the inaccuracy of the current water disaster monitoring method for coal mines.

[0005] For the above problems, there is no good solution at present. Summary of the Invention

[0006] The embodiments of the present application provide a coal mine disaster identification method based on a deep learning algorithm and related devices, which are used to improve the accuracy of water disaster monitoring for coal mines.

[0007] To achieve the above object, the embodiments of the present application adopt the following technical solutions:

[0008] In a first aspect, a coal mine disaster identification method based on a deep learning algorithm is provided, and the method includes:

[0009] Obtain the laser scanning point cloud data, geological drilling data, and geophysical exploration data of the coal mine;

[0010] Divide the target area including the coal mine into a plurality of three-dimensional grid cells, and interpolate the laser scanning point cloud data, the geological drilling data, and the geophysical exploration data into the corresponding three-dimensional grid cells to obtain a first three-dimensional geological model;

[0011] Add formation parameters and coal seam parameters to the first 3D geological model to obtain a second 3D geological model;

[0012] Associate the second 3D geological model with a pre - constructed big data model, obtain the current season, and acquire the monitoring data for the current season according to the big data model. The monitoring data includes weather parameters, water level data, and environmental data;

[0013] Extract features from the current season, the weather parameters, the water level data, and the environmental data to obtain a multi - dimensional time series;

[0014] Input the multi - dimensional time series into a pre - constructed deep learning model to obtain water level prediction values for each time unit in a preset future time period;

[0015] Input the water level prediction values for each time unit in the preset future time period into the big data model to correct the data of the water level prediction values for each time unit in the preset future time period;

[0016] Perform spatial combination of the corrected water level prediction values with the second 3D geological model to obtain a dynamic 3D geological model including water level changes in a preset future time period;

[0017] Based on the dynamic 3D geological model, determine the probability of water disaster occurrence in the coal mine shaft.

[0018] In a possible implementation manner of the first aspect, the dividing the target area including the coal mine shaft into multiple 3D grid units and interpolating the laser scanning point cloud data, the geological drilling data, and the geophysical exploration data into the corresponding 3D grid units to obtain the first 3D geological model includes:

[0019] Obtain the area of the target area, and determine the density of the laser scanning point cloud data according to the laser scanning point cloud data;

[0020] Determine the grid resolution according to the area of the target area and the density of the laser scanning point cloud data;

[0021] Divide the target area into multiple 3D grid units according to the grid resolution;

[0022] In the case where any one of the 3D grid units contains the laser scanning point cloud data, the geological drilling data, and / or the geophysical exploration data, interpolate the laser scanning point cloud data, the geological drilling data, and / or the geophysical exploration data into the 3D grid unit.

[0023] In another possible implementation of the first aspect, determining the grid resolution according to the area of the target region and the density of the laser scanning point cloud data includes:

[0024] Taking a region with a preset area as a basic unit, and dividing the target region into multiple sub-regions according to the basic unit;

[0025] Calculating the average density and standard deviation of the laser scanning point cloud data in each of the sub-regions;

[0026] Using a preset adaptive calculation formula to calculate the grid resolution of each of the sub-regions according to the average density and the standard deviation, wherein, when the grid resolution is less than a first preset value, setting the grid resolution to the first preset value, and when the grid resolution is greater than a second preset value, setting the grid resolution to the second preset value, and the second preset value is greater than the first preset value.

[0027] In another possible implementation of the first aspect, associating the second three-dimensional geological model with a pre-constructed big data model, obtaining the current season, and obtaining the monitoring data of the current season according to the big data model includes:

[0028] Constructing an octree spatial index structure for the current season, the octree spatial index structure including a plurality of cubic sub-spaces and a plurality of nodes storing boundary coordinates and monitoring data;

[0029] Calculating the center point coordinates of each of the three-dimensional grid cells in the second three-dimensional geological model;

[0030] Querying, according to the boundary coordinates, a target cubic sub-space including the center point coordinates in the octree spatial index structure, and determining a node corresponding to the center point coordinates in the target cubic sub-space, and the monitoring data corresponding to the center point coordinates is the monitoring data stored in the node.

[0031] In another possible implementation of the first aspect, the deep learning model is a long short-term memory network, the long short-term memory network includes an input layer, a first LSTM layer, a second LSTM layer, a third LSTM layer, a fully connected layer, and an output layer, and inputting the multi-dimensional time series into a pre-constructed deep learning model to obtain water level prediction values for each time unit in a preset future time period includes:

[0032] Inputting the multi-dimensional time series through the input layer, and respectively passing through the first LSTM layer, the second LSTM layer, the third LSTM layer, and the fully connected layer, and obtaining water level prediction values for each time unit in a preset future time period through the output layer.

[0033] In another possible implementation of the first aspect, inputting the predicted water level values of each time unit in the future preset time period into the big data model to perform data correction on the predicted water level values of each time unit in the future preset time period includes:

[0034] Determine the seasonal index, upper water level value, lower water level value, and water level trend of the current season according to the big data model;

[0035] Smooth the predicted water level values according to the seasonal index, the upper water level value, and the lower water level value;

[0036] Perform trend correction on the smoothed predicted water level values according to the water level trend to obtain the predicted water level values after data correction.

[0037] In another possible implementation of the first aspect, spatially combining the corrected predicted water level values with the second three-dimensional geological model to obtain a dynamic three-dimensional geological model including water level changes in the future preset time period includes:

[0038] Define the boundary and initial water level surface of the water body in the second three-dimensional geological model;

[0039] Calculate the elevation of the water level surface at each time point according to the corrected predicted water level values;

[0040] Adopt a preset bilinear difference interpolation algorithm to spread the elevation of the water level surface at each time point to a preset three-dimensional water level model to obtain a water level surface grid model;

[0041] Fuse the water level surface grid model with the second three-dimensional geological model to obtain a fusion model;

[0042] In the fusion model, construct a three-dimensional model for each time point, and according to the three-dimensional models of each time point, obtain a dynamic three-dimensional geological model including water level changes in the future preset time period.

[0043] In another possible implementation of the first aspect, determining the probability of water disaster occurring in the coal mine based on the dynamic three-dimensional geological model includes:

[0044] Calculate the key parameters corresponding to the three-dimensional model of each time point, where the key parameters include the volume ratio of the dangerous area, the water level rising speed, and the carrying capacity of the drainage system;

[0045] Adopt a preset water disaster calculation formula to calculate the probability of water disaster occurrence according to the key parameters.

[0046] In a second aspect, the present application provides a coal mine disaster recognition device based on a deep learning algorithm, including:

[0047] a memory configured to store instructions; and

[0048] a processor configured to call the instructions from the memory and capable of implementing the above-mentioned coal mine disaster recognition method based on a deep learning algorithm when executing the instructions.

[0049] In a third aspect, the present application provides an electronic device, including:

[0050] the above-mentioned coal mine disaster recognition device based on a deep learning algorithm.

[0051] Through the above technical solutions, by adding stratum parameters and coal seam parameters to the first three-dimensional geological model, a second three-dimensional geological model is obtained, which can more accurately reflect the actual geological conditions of the coal mine; the second three-dimensional geological model is associated with the big data model, and the monitoring data of the current season is obtained by using the big data model, thereby taking into account the seasonal changes of the coal mine; using the pre-constructed deep learning model for water level prediction can improve the prediction accuracy, and using the big data model to correct the water level prediction value can further improve the prediction accuracy; spatially combining the corrected water level prediction value with the second three-dimensional geological model to obtain a dynamic three-dimensional geological model can more intuitively display the future water level changes, and the probability of water disaster occurrence can be evaluated based on the dynamic three-dimensional geological model, thereby effectively improving the accuracy of water disaster monitoring of the coal mine. In summary, by taking into account the seasonal changes of the coal mine, the problem that seasonal changes mask sudden water conditions is effectively solved. At the same time, by using the deep learning model and the big data model, the accuracy of water disaster monitoring of the coal mine can be effectively improved.

[0052] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent specific implementation part. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a schematic flowchart of a coal mine disaster recognition method based on a deep learning algorithm provided by an embodiment of the present application;

[0054] Figure 2 is a schematic diagram of a target area and three-dimensional grid cells provided by an embodiment of the present application;

[0055] Figure 3 is a schematic structural diagram of a deep learning model provided by an embodiment of the present application.

[0056] DESCRIPTION OF THE REFERENCE NUMERALS

[0057] 1 Target area 10 Three-dimensional grid cells Detailed implementation manners

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. It should be understood that the specific implementation manners described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts fall within the scope of protection of the present application.

[0059] It should be noted that if there are directional indications (such as up, down, left, right, front, back...) involved in the embodiments of the present application, the directional indications are only used to explain the relative positional relationship and movement conditions between components in a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indications will also change accordingly.

[0060] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In addition, the technical solutions between various embodiments can be combined with each other, but it must be based on the fact that those of ordinary skill in the art can implement them. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present application.

[0061] Figure 1 A schematic flow diagram of a coal mine disaster identification method based on a deep learning algorithm according to an embodiment of the present application is schematically shown. As Figure 1 shown, an embodiment of the present application provides a coal mine disaster identification method based on a deep learning algorithm, and the method may include the following steps.

[0062] S101. Obtain the laser scanning point cloud data, geological drilling data, and geophysical exploration data of the coal mine;

[0063] S102. Divide the target area including the coal mine into multiple three-dimensional grid cells, and interpolate the laser scanning point cloud data, geological drilling data, and geophysical exploration data into the corresponding three-dimensional grid cells to obtain a first three-dimensional geological model;

[0064] S103. Add stratum parameters and coal seam parameters to the first three-dimensional geological model to obtain a second three-dimensional geological model;

[0065] S104. Associate the second 3D geological model with the pre-constructed big data model, obtain the current season, and acquire the monitoring data for the current season according to the big data model. The monitoring data includes weather parameters, water level data, and environmental data.

[0066] S105. Extract features from the current season, weather parameters, water level data, and environmental data to obtain a multi-dimensional time series.

[0067] S106. Input the multi-dimensional time series into the pre-constructed deep learning model to obtain the water level prediction values for each time unit in a preset future time period.

[0068] S107. Input the water level prediction values for each time unit in the preset future time period into the big data model to correct the data of the water level prediction values for each time unit in the preset future time period.

[0069] S108. Combine the corrected water level prediction values with the second 3D geological model spatially to obtain a dynamic 3D geological model including water level changes in a preset future time period.

[0070] S109. Determine the probability of water damage occurring in the coal mine shaft based on the dynamic 3D geological model.

[0071] First, obtain the laser scanning point cloud data, geological drilling data, and geophysical exploration data of the coal mine shaft. Specifically, the laser scanning point cloud data includes underground roadway structure data, shaft structure data, and chamber structure data.

[0072] In specific implementation, the laser scanning point cloud data consists of discrete three-dimensional coordinate points and is used to describe the geometric shape of the coal mine shaft. The laser scanning point cloud data can be obtained by a laser rangefinder. The laser rangefinder can measure the three-dimensional coordinates of the object surface by emitting a laser beam and receiving the reflected signal to accurately capture the underground roadway structure data, shaft structure data, and chamber structure data. Among them, the underground roadway structure data includes the width, height, length, and spatial distribution of the roadway; the shaft structure data includes the location, diameter, and shape of the shaft; the chamber structure data includes the size and spatial layout of the chamber.

[0073] The geological drilling data can be obtained by drilling and sampling around the coal mine shaft and includes information such as the distribution, thickness, dip angle, etc. of coal seams and various rock layers, and also includes the physical properties of the rock layers, such as density, porosity, etc.

[0074] The geophysical exploration data can obtain underground structure information through geophysical exploration methods, such as seismic wave detection, resistivity measurement, etc. In addition, in this embodiment, the historical construction data of the coal mine shaft can also be obtained, and the mining history, support method, drainage system, etc. can be obtained through the historical construction data of the coal mine shaft.

[0075] Through the above multi-source data collection, it can help identify potential water hazard risk areas. For example, through laser scanning point cloud data, it can be found that the width of a certain roadway gradually narrows from 3 meters at the entrance to 2.5 meters, with a length of 500 meters and a height maintained at 2.8 meters; while geological drilling data may show that there is a water-bearing sandstone layer with a thickness of 2 meters 50 meters above this roadway.

[0076] Divide the target area 1 containing the coal mine shaft into multiple three-dimensional grid cells 10, and each three-dimensional grid cell 10 represents a unit volume in space. Figure 2 The figure shows a schematic diagram of a target area and three-dimensional grid cells provided by an embodiment of the present application, as Figure 2 shown. In the case where the target area 1 is a cube, the three-dimensional grid cell 10 is a cube grid cell.

[0077] In this embodiment, spatial interpolation techniques can be used to interpolate the laser scanning point cloud data, geological drilling data, and geophysical exploration data into the corresponding three-dimensional grid cells respectively. For the laser scanning point cloud data, the nearest neighbor interpolation method can be used to assign the laser scanning point cloud data to the grid.

[0078] Since geological drilling data are usually discrete point data, Kriging interpolation can be used to extend it in space. Geophysical exploration data can be three-dimensionally interpolated using geostatistical methods. For example, assume that the target area 1 including the coal mine shaft is 100m x 100m x 50m, and it can be divided into 1m x 1m x 1m cube grid cells (three-dimensional grid cells 10), a total of 500,000 cells. Then, corresponding attribute values, such as lithology codes, coal property codes, etc., are assigned to each three-dimensional grid cell, and the laser scanning point cloud data, geological drilling data, and geophysical exploration data are respectively interpolated into the corresponding three-dimensional grid cells 10, and finally the obtained first three-dimensional geological model, where the first three-dimensional geological model is a voxel model containing various geological attributes.

[0079] Adding rock layer parameters and coal seam parameters to the first three-dimensional geological model can enhance the ability of the first three-dimensional geological model to characterize geological and physical properties. Specifically, the rock layer parameters include permeability coefficient and strength parameters, and the coal seam parameters include thickness parameter, dip parameter, and burial depth parameter.

[0080] Among them, the permeability coefficient in the rock layer parameters describes the permeability of the rock to fluids, usually in units of Darcy (D) or centimeters per second (cm / s). For example, the permeability coefficient of sandstone may be in the range of 10 -3 to 10 -5 cm / s, while that of shale may be as low as 10 -8cm / s. The strength parameters include the uniaxial compressive strength, tensile strength, cohesion, and internal friction angle of the rock, etc. The coal seam parameters include thickness parameter, dip angle parameter, and burial depth parameter. The uniaxial compressive strength of the rock can be determined through rock mechanics tests, while the permeability coefficient can be measured through hydraulic fracturing tests.

[0081] For each three-dimensional grid cell, the corresponding rock layer parameters can be added according to its corresponding lithology; similarly, the corresponding coal seam parameters can be added according to its corresponding coal property.

[0082] After that, the second three-dimensional geological model is associated with the pre-constructed big data model. The pre-constructed big data model includes multi-source data such as historical geological data, historical disaster records, and environmental monitoring data. The association process first needs to establish a spatial correspondence relationship between the second three-dimensional geological model and the pre-constructed big data model to ensure that each three-dimensional grid cell in the second three-dimensional geological model can accurately correspond to the geographical location in the big data model. In specific implementation, it can be achieved through spatial indexing technology. For example, a spatial index structure can be used to accelerate the spatial query and matching process. Through the association, the second three-dimensional geological model can better capture regional geological features and long-term environmental change trends, thereby improving the accuracy and reliability of disaster prediction.

[0083] Obtain the current season and obtain the corresponding monitoring data according to the big data model. In this embodiment, the current season can be achieved through the system clock or user input. According to the determined season, the corresponding monitoring data is extracted from the big data model. The weather parameters include temperature, precipitation, humidity, wind speed, etc. For example, the daily average temperature, daily cumulative precipitation, and daily variation of relative humidity can be obtained. The water level data includes surface water level and groundwater level. Among them, the surface water level includes the water level changes of rivers and lakes near the coal mine, while the groundwater level can be obtained through observation wells. The environmental data includes soil moisture content, surface runoff, vegetation coverage, and ground subsidence, and the environmental data can be obtained from the geographic information database.

[0084] Extract features from the current season, weather parameters, water level data, and environmental data to obtain a multi-dimensional time series. For seasonal information, one-hot encoding can be used to convert it into numerical features. For example, spring can be encoded as [1, 0, 0, 0], summer as [0, 1, 0, 0], and so on. For weather parameters, statistical features such as moving averages and standard deviations can be calculated to capture short-term and long-term weather change trends. For example, the average temperature and the standard deviation of precipitation over the past 7 days can be calculated. Water level data can extract features such as trends, periodicity, and autocorrelation, such as using wavelet transform to decompose water level changes at different scales. For soil moisture content in environmental data, a saturation index can be calculated, for surface runoff, a base flow separation index can be extracted, and for vegetation coverage, a growing season index can be calculated. Among them, the current season can be represented by digital encoding as a periodic time series, and the weather parameters, water level data, and environmental data are all time series data. Therefore, features can be extracted from the current season, weather parameters, water level data, and environmental data to obtain a multi-dimensional time series.

[0085] It should be noted that during the process of feature extraction from the current season, weather parameters, water level data, and environmental data, it is necessary to ensure that all data are aligned in time so that all sequences have the same time interval.

[0086] Next, input the multi-dimensional time series into a pre-built deep learning model, and the water level prediction values for each time unit in a preset future time period can be obtained.

[0087] Among them, the pre-built deep learning model can effectively capture long-term dependencies and non-linear patterns in time series data. During the model training process of this embodiment, the Adam optimizer is used, the learning rate is set to 0.001, the batch size is 64, and the number of training epochs is 1000. The loss function is the root mean square error. In the prediction stage, the deep learning model adopts a sliding window method. Each time, a time series of a fixed length (such as the past 30 days) is input to predict the water level for a future period (such as the next 7 days). Then the sliding window is moved, and some of the prediction results are used as new inputs to continue predicting the next time period until the entire preset time period is covered. Finally, the water level prediction values for each time unit (such as daily or hourly) within the preset future time period are output.

[0088] Input the water level prediction values for each time unit in the preset future time period into a big data model for data correction, and the water level trend of the water level prediction values can be corrected according to the water level trend of the current season in the big data model.

[0089] The corrected water level prediction values are spatially combined with the second three-dimensional geological model to obtain a dynamic three-dimensional geological model including water level changes for a preset future time period. Specifically, first, the corrected water level prediction data is converted into a format suitable for three-dimensional spatial representation. For example, interpolation algorithms are used to convert discrete water level data into a continuous water level surface. Then, the water level surface data is spatially superimposed and fused with the second three-dimensional geological model.

[0090] In one embodiment, it can be implemented using spatial analysis software such as ArcGIS or QGIS in combination with three-dimensional modeling tools such as Gocad or Petrel. The specific steps may include: (1) Import the water level prediction data into ArcGIS to create a water level raster; (2) Register the water level raster and the second three-dimensional geological model in the same coordinate system; (3) Fuse the water level raster and the second three-dimensional geological model in Gocad to obtain a dynamic three-dimensional geological model.

[0091] The dynamic three-dimensional geological model not only contains static geological structure information but also incorporates water level dynamic information that changes over time, and can intuitively display the impact of water level changes on the coal mine shaft.

[0092] Finally, based on the dynamic three-dimensional geological model, the probability of water disaster occurrence in the coal mine shaft is determined. First, key risk factors need to be extracted from the dynamic three-dimensional geological model, such as the volume ratio of the dangerous area, the water level rising speed, the rock layer strength, the fault location, and the bearing capacity of the drainage system, etc. According to the risk factors, the probability of water disaster occurrence in the coal mine shaft is calculated using a preset calculation formula.

[0093] In this embodiment, rock layer parameters and coal seam parameters are added to the first three-dimensional geological model to obtain the second three-dimensional geological model, which can more accurately reflect the actual geological conditions of the coal mine shaft; the second three-dimensional geological model is associated with the big data model, and the monitoring data of the current season is obtained using the big data model, thus taking into account the seasonal changes of the coal mine shaft; using a pre-constructed deep learning model for water level prediction can improve the prediction accuracy, and using the big data model to correct the water level prediction values can further improve the prediction accuracy; the corrected water level prediction values are spatially combined with the second three-dimensional geological model to obtain a dynamic three-dimensional geological model, which can more intuitively display future water level changes, and the probability of water disaster occurrence can be evaluated based on the dynamic three-dimensional geological model, thereby effectively improving the accuracy of water disaster monitoring of the coal mine shaft. In summary, by taking into account the seasonal changes of the coal mine shaft, the problem that seasonal changes mask sudden water conditions is effectively solved. At the same time, by using the deep learning model and the big data model, the accuracy of water disaster monitoring of the coal mine shaft can be effectively improved.

[0094] In one implementation of this embodiment, the target area including the coal mine shaft is divided into multiple three-dimensional grid cells, and the laser scanning point cloud data, geological drilling data, and geophysical exploration data are interpolated into the corresponding three-dimensional grid cells to obtain the first three-dimensional geological model, including the following steps:

[0095] S201. Obtain the area of the target area, and determine the density of the laser scanning point cloud data according to the laser scanning point cloud data;

[0096] S202. Determine the grid resolution according to the area of the target area and the density of the laser scanning point cloud data;

[0097] S203. Divide the target area into multiple three-dimensional grid cells according to the grid resolution;

[0098] S204. In the case where any three-dimensional grid cell contains laser scanning point cloud data, geological drilling data, and / or geophysical exploration data, interpolate the laser scanning point cloud data, geological drilling data, and / or geophysical exploration data into the three-dimensional grid cell.

[0099] First, obtain the area of the target area, which can be completed through surveying and mapping or a GIS system. The unit of area can be square meters or square kilometers. Next, determine its density according to the laser scanning point cloud data. The density of the laser scanning point cloud data is usually expressed as the number of points per square meter and can be obtained by calculating the total number of points divided by the covered area. For example, if there are 80 million points in an area of 800,000 square meters, then the average density is 100 points per square meter.

[0100] Determine the grid resolution according to the area of the target area and the density of the laser scanning point cloud data. Generally speaking, the grid resolution should be less than or equal to the average point spacing of the point cloud data to ensure that important terrain details are not lost. The following formula can be used to estimate the appropriate grid resolution:

[0101]

[0102] For example, if the point cloud density is 100 points per square meter, then the grid resolution is approximately 0.1 meter. However, in specific implementations, each grid resolution may be different. If there is a fault with a width of 5 meters in the target area, then the grid resolution should be much less than 5 meters to accurately represent the fault.

[0103] After that, the target area is divided into multiple three-dimensional grid cells according to the determined grid resolution. First, it is necessary to determine the spatial range of the entire model, including the minimum and maximum coordinate values in the X, Y, and Z directions. Then, according to the previously determined grid resolution, equidistant division is carried out in each direction. For example, if the range in the X direction is 0 - 1000 meters and the grid resolution is 1 meter, then there will be 1000 grids in the X direction. The Y and Z directions are processed similarly. In this way, the entire target area is divided into a large number of cubic or cuboid grid cells. Each grid cell has a unique spatial coordinate and index, which is convenient for subsequent data interpolation and attribute assignment. In specific implementation, in order to improve the calculation efficiency, higher-resolution grids can be used in key areas, and lower-resolution grids can be used in areas with little change, that is, adaptive grid division is performed.

[0104] In the case where any one of the three-dimensional grid cells contains laser scanning point cloud data, geological drilling data, and / or geophysical exploration data, the laser scanning point cloud data, geological drilling data, and / or geophysical exploration data are interpolated into the three-dimensional grid cells.

[0105] In this embodiment, spatial interpolation techniques can be used to interpolate the laser scanning point cloud data, geological drilling data, and geophysical exploration data into the corresponding three-dimensional grid cells respectively. For the laser scanning point cloud data, the nearest neighbor interpolation method can be used to allocate the laser scanning point cloud data to the grids.

[0106] Since geological drilling data is usually discrete point data, Kriging interpolation method can be used to extend it in space. Geophysical exploration data can be interpolated three-dimensionally using geostatistical methods. For example, assuming that the target area including a coal mine shaft is 100m x 100m x 50m, it can be divided into 1m x 1m x 1m cubic grid cells (three-dimensional grid cells), a total of 500,000 cells. Then, corresponding attribute values, such as lithology codes, coal property codes, etc., are assigned to each cell, and the laser scanning point cloud data, geological drilling data, and geophysical exploration data are interpolated into the corresponding three-dimensional grid cells respectively. Finally, the obtained first three-dimensional geological model, where the first three-dimensional geological model is a voxel model containing various geological attributes.

[0107] The first three-dimensional geological model constructed in this embodiment integrates the laser scanning point cloud data, geological drilling data, and geophysical exploration data, and can accurately reflect the geological characteristics and spatial distribution of the target area. By interpolating the laser scanning point cloud data, geological drilling data, and / or geophysical exploration data into the three-dimensional grid cells, the accuracy of the first three-dimensional geological model is ensured.

[0108] In one implementation manner of this embodiment, the grid resolution is determined according to the area of the target area and the density of the laser scanning point cloud data, including the following steps:

[0109] S301. Take the area of a preset size as a basic unit, and divide the target area into multiple sub - areas according to the basic unit;

[0110] S302. Calculate the average density and standard deviation of the laser scanning point cloud data in each sub - area;

[0111] S303. Use a preset adaptive calculation formula to calculate the grid resolution of each sub - area according to the average density and standard deviation. Wherein, when the grid resolution is less than the first preset value, set the grid resolution to the first preset value, and when the grid resolution is greater than the second preset value, set the grid resolution to the second preset value, and the second preset value is greater than the first preset value.

[0112] First, take the area of a preset size as a basic unit, and divide the target area into multiple sub - areas according to the basic unit. The basic unit can be determined based on the total area of the target area and the terrain complexity. For example, for a target area with a total area of 10 square kilometers, the basic unit can be an area of 100 meters×100 meters. In this way, the entire target area is divided into 1000 sub - areas.

[0113] In one embodiment, GIS (Geographic Information System) software can be used for the division. A regular grid can be created to cover the entire target area, and the area of each grid cell is the preset basic unit area. If the shape of the target area is irregular, there may be some grid cells that only partially cover the target area. In this case, those grid cells that cover more than 50% of the target area can be selected to be retained.

[0114] In another embodiment, if there are obvious geographical or geological boundaries (such as rivers, faults) within the target area, the above - mentioned grid division may ignore the natural boundaries. In this case, the average density and standard deviation of the laser scanning point cloud data in each sub - area can be calculated. The average density and standard deviation of the laser scanning point cloud data can determine the distribution characteristics of the three - dimensional scanning point cloud data.

[0115] Specifically, first, for each sub - area, it is necessary to determine all the point cloud data points that fall within the area. This can usually be determined by checking whether the coordinates of each point are within the boundaries of the sub - area. Divide the total number of point cloud data points within the sub - area by the area of the sub - area to obtain the average density of the laser scanning point cloud data.

[0116] The steps for calculating the standard deviation are as follows: further divide the sub - area into smaller grids (for example, 1 - meter×1 - meter small grids); calculate the number of points in each small grid to obtain multiple density samples; use the standard deviation calculation formula to calculate the standard deviation of the laser scanning point cloud data.

[0117] The calculation formula for the standard deviation is as follows:

[0118]

[0119] where x is the density of each small grid, μ is the average density, and n is the number of small grids.

[0120] In actual operation, some sub-regions may contain very few or no laser scanning point cloud data, which may be caused by occlusion during the scanning process or incomplete data acquisition. In this case, interpolation or estimation is required to fill in the laser scanning point cloud data for these sub-regions.

[0121] Using a preset adaptive calculation formula, the grid resolution of each sub-region is calculated based on the average density and the standard deviation.

[0122] The adaptive calculation formula is as follows:

[0123]

[0124] where k is a scaling factor used to adjust the overall resolution level; α is a preset weight coefficient used to control the influence degree of the standard deviation on the resolution. In the adaptive calculation formula, the resolution is inversely proportional to the point cloud density, and at the same time, the non-uniformity of the density distribution is reflected through the standard deviation.

[0125] For example, assume that the average density of a certain sub-region is 100 points per square meter, the standard deviation is 20 points per square meter, k = 0.5, and α = 0.1. Then the grid resolution of this sub-region is calculated as follows:

[0126]

[0127] However, in this embodiment, in order to avoid the grid resolution being too small or too large, upper and lower limit thresholds are set. If the calculated grid resolution is less than the first preset value (for example, 0.05 meters), the grid resolution is set to the first preset value. This is used to prevent generating too many grids in high-density regions, resulting in an excessive calculation burden. Similarly, if the calculated grid resolution is greater than the second preset value (for example, 1 meter), the grid resolution is set to the second preset value. This is used to ensure that a certain model accuracy can still be maintained in low-density regions.

[0128] In this embodiment, the target area is first divided into multiple sub-areas, and then the laser scanning point cloud data of each sub-area is statistically analyzed to obtain the average density and standard deviation. Finally, the grid resolution is dynamically adjusted according to the average density and standard deviation, so that the grid design can be optimized according to the actual distribution characteristics of the laser scanning point cloud data, improving the calculation efficiency while ensuring the model accuracy. By setting upper and lower threshold values, extreme grid sizes can also be avoided, ensuring the overall quality of the model. The above method for adaptively adjusting the grid resolution can achieve the best balance between accuracy and efficiency among different sub-areas.

[0129] In one implementation of this embodiment, the second three-dimensional geological model is associated with a pre-constructed big data model, and the current season is obtained. According to the big data model, the monitoring data for the current season is obtained, including the following steps:

[0130] S401. Construct an octree spatial index structure for the current season. The octree spatial index structure includes multiple cubic sub-spaces and multiple nodes storing boundary coordinates and monitoring data;

[0131] S402. Calculate the center point coordinates of each three-dimensional grid cell in the second three-dimensional geological model;

[0132] S403. According to the boundary coordinates, query the target cubic sub-space including the center point coordinates in the octree spatial index structure, and determine the node corresponding to the center point coordinates in the target cubic sub-space. The monitoring data corresponding to the center point coordinates is the monitoring data stored in the node.

[0133] The octree spatial index structure is a tree-shaped data structure, where each node has at most eight child nodes. In three-dimensional space, it organizes data by recursively dividing the space into eight equal cubic sub-spaces.

[0134] The construction process of the octree spatial index structure starts from a root cube containing the entire target area. Then, according to predefined conditions (such as data density), the root cube is recursively subdivided into eight smaller sub-cubes. Each sub-cube becomes a new node in the octree. The above process is executed cyclically until a preset termination condition is reached, such as the minimum cube size or the maximum tree depth.

[0135] At each node, two types of key information are stored: boundary coordinates and monitoring data. The boundary coordinates define the range of the cubic subspace represented by the node, usually including six values: xmin, xmax, ymin, ymax, zmin, zmax. For example, assume the target area is a cube with a side length of 1000 meters. The first-level division will generate 8 sub-cubes with a side length of 500 meters. If one of the sub-cubes contains more data points than the preset threshold (such as 1000 points), then this sub-cube is further divided into 8 smaller sub-cubes with a side length of 250 meters. The above process is executed cyclically until all sub-cubes meet the termination condition or no longer contain any data points.

[0136] The octree spatial index structure can adaptively adjust the fineness of the spatial division according to the distribution characteristics of the data. In data-dense areas, the octree will have deeper levels and smaller sub-cubes, while in data-sparse areas, there will be larger sub-cubes, enabling the octree to greatly reduce storage space and improve query efficiency while maintaining high precision.

[0137] Next, calculate the center point coordinates of each three-dimensional grid cell in the second three-dimensional geological model. Specifically, in three-dimensional space, each three-dimensional grid cell is usually a hexahedron. The basic principle for calculating the center point coordinates is to determine the geometric center of the hexahedron. For a regular cuboid, its center point coordinates can be calculated as follows: x_center = (x_min + x_max) / 2; y_center = (y_min + y_max) / 2; z_center = (z_min + z_max) / 2. Where x_min, x_max, y_min, y_max, z_min, z_max are the minimum and maximum coordinate values of the grid cell in the x, y, and z directions respectively.

[0138] Finally, according to the boundary coordinates, query the target cubic subspace including the center point coordinates in the octree spatial index structure, and determine the node corresponding to the center point coordinates in the target cubic subspace. In this embodiment, starting from the root node of the octree, compare the center point coordinates with the boundaries of the cubic subspace represented by the current node. If the center point falls within the current cubic subspace, continue to check the eight child nodes of the current node until reaching the leaf node or finding the smallest subspace containing the node.

[0139] For example, assume there is a grid cell with center point coordinates (105, 210, 315). The query process can be as follows:

[0140] 1. First, check the root node. Assume the spatial range it represents is (0 - 1000, 0 - 1000, 0 - 1000), and the center node is within this range.

[0141] 2. Check the 8 child nodes of the root node. Assume that a child node with a range of (0 - 500, 0 - 500, 0 - 500) containing the central node is found.

[0142] 3. Continue to check the child nodes of the child node with a range of (0 - 500, 0 - 500, 0 - 500). It is possible to find a child node with a range of (0 - 250, 125 - 375, 250 - 500).

[0143] 4. If the child node with a range of (0 - 250, 125 - 375, 250 - 500) is the smallest subspace containing the central node, the query stops.

[0144] After determining the target cube subspace, the monitoring data stored in the node corresponding to the target cube subspace (the node corresponding to the central point coordinates) can be accessed.

[0145] In this embodiment, an octree spatial index structure is first constructed. Then, the central point coordinates of each grid cell are calculated to provide a basis for data matching. Finally, spatial queries are performed using the octree, realizing the precise correspondence between the second three-dimensional geological model and the monitoring data, effectively improving the efficiency of data retrieval and matching, and thus significantly enhancing the accuracy of the second three-dimensional geological model.

[0146] In one implementation of this embodiment, the deep learning model is a long short-term memory network. The long short-term memory network includes an input layer, a first LSTM layer, a second LSTM layer, a third LSTM layer, a fully connected layer, and an output layer. Inputting the multi-dimensional time series into the pre-constructed deep learning model to obtain the water level prediction values for each time unit in a future preset time period includes the following steps:

[0147] S501. Input the multi-dimensional time series through the input layer, and pass through the first LSTM layer, the second LSTM layer, the third LSTM layer, and the fully connected layer respectively, and obtain the water level prediction values for each time unit in a future preset time period through the output layer.

[0148] Figure 3 Shows a schematic structural diagram of a deep learning model provided by an embodiment of the present application. As Figure 3 shown, the deep learning model includes an input layer, a first LSTM layer, a second LSTM layer, a third LSTM layer, a fully connected layer, and an output layer.

[0149] The multi-dimensional time series is input through the input layer and passes through the first LSTM layer, the second LSTM layer, the third LSTM layer, and the fully connected layer respectively, and finally the water level prediction values of each time unit in the preset future time period are obtained through the output layer. The long short-term memory network can be used to capture the long-term dependencies in the time series data, thereby achieving accurate prediction of the future water level.

[0150] First, the multi-dimensional time series data enters the network through the input layer. The input layer is used to convert the multi-dimensional time series data into a format suitable for network processing. For example, if there are n features, the input for each time step may be an n-dimensional vector. Assuming that the data of the past 30 days is used to predict the future, the input may be a matrix with a shape of (30, n).

[0151] Next, the multi-dimensional time series data passes through three LSTM layers in sequence. Each LSTM layer contains multiple LSTM cells, and the LSTM cells can learn and remember long-term dependencies. The LSTM cells include a forget gate, an input gate, and an output gate to capture the long-term patterns in the time series.

[0152] Taking the first LSTM layer as an example, its mathematical expression can be simplified as:

[0153] ft = σ(Wf · [ht-1, xt] + bf) / / Forget gate;

[0154] / / Input gate;

[0155] / / Output gate;

[0156] / / Cell state update;

[0157] ht = ot * tanh(ct) / / Hidden state output.

[0158] Where σ is the sigmoid function, * represents the Hadamard product, · represents matrix multiplication, xt is the input at the current time step, ht-1 is the hidden state at the previous time step, and ct-1 is the cell state at the previous time step.

[0159] The working principles of the second and third LSTM layers are similar, but their inputs are the outputs of the previous layer. The stacking of multiple LSTM layers enables the deep learning network to learn more complex time dependencies.

[0160] After the LSTM layer, the data passes through the fully connected layer. The fully connected layer is used to map the features extracted by the LSTM layer to the final prediction values. It can be expressed as:

[0161] y = f(Wx + b);

[0162] Among them, x is the input vector, W is the weight matrix, b is the bias vector, and f is the activation function (such as ReLU).

[0163] Finally, the final water level prediction value is obtained through the output layer. The output layer usually uses a linear activation function to directly output the predicted water level value. In this embodiment, the long short-term memory network is trained through the backpropagation algorithm, and the goal is to minimize the error between the predicted value and the actual water level.

[0164] By adopting a multi-layer LSTM network structure, this embodiment can effectively handle the complex time-dependent relationships in water level prediction. The input layer receives multi-dimensional time series data, and the three LSTM layers gradually extract and learn the long-term and short-term patterns in the time series. The fully connected layer performs feature integration, and finally the output layer gives the specific water level prediction value, thereby providing a more accurate water level prediction.

[0165] In one implementation of this embodiment, the water level prediction values of each time unit in a preset future time period are input into the big data model to correct the data of the water level prediction values of each time unit in the preset future time period, including the following steps:

[0166] S601. According to the big data model, determine the seasonal index, water level upper limit value, water level lower limit value, and water level trend of the current season;

[0167] S602. Smooth the water level prediction values according to the seasonal index, water level upper limit value, and water level lower limit value;

[0168] S603. According to the water level trend, correct the smoothed water level prediction values to obtain the data-corrected water level prediction values.

[0169] Determine the seasonal index, water level upper limit value, water level lower limit value, and water level trend of the current season according to the big data model. Specifically, the seasonal index reflects the periodic change pattern of the water level in different seasons. It is calculated by decomposing the time series, and the original data is decomposed into trend, seasonal, and random components. For example, for monthly data, the 12-month centered moving average can be calculated, and then the original data is divided by this moving average to obtain the seasonal factor. Repeat this process for multiple years of data, and then take the average of the seasonal factors for each month to obtain the seasonal index.

[0170] The water level upper limit value and the lower limit value represent the reasonable range of the water level in a specific season. It can be determined by analyzing the distribution of historical data. For example, the 95th percentile can be used as the upper limit and the 5th percentile can be used as the lower limit, which can exclude the influence of extreme outliers while retaining most of the normal fluctuations.

[0171] The water level trend reflects the long-term change direction and can be determined by linear regression. For example, when using linear regression, time can be used as the independent variable and the water level as the dependent variable to fit a straight line. The slope of the straight line represents the water level trend.

[0172] Smoothing the water level prediction value according to the seasonal index, the upper limit value of the water level, and the lower limit value of the water level can reduce abnormal fluctuations in the prediction and make the prediction result more in line with the seasonal law and reasonable range.

[0173] A possible implementation method is as follows: First, divide the original prediction value by the corresponding seasonal index to obtain the deseasonalized value: deseasonalized value = original prediction value / seasonal index; then, smooth the deseasonalized value. Use the simple moving average method: smoothed value = (sum of the previous n deseasonalized values) / n, where n can be an odd number between 3 and 7; next, multiply the smoothed value by the seasonal index to restore the seasonal characteristics: smoothed value with restored seasonality = smoothed value * seasonal index; finally, check whether the smoothed value with restored seasonality is between the upper limit value and the lower limit value of the water level. If it exceeds the range, limit it to the nearest boundary value.

[0174] According to the water level trend, perform trend correction on the smoothed water level prediction value to obtain the water level prediction value after data correction. The steps are as follows: First, calculate the difference between the smoothed prediction value and the trend line: difference = smoothed prediction value - trend line prediction value, where the trend line prediction value can be the water level trend determined above. Then, adjust the smoothed prediction value according to the magnitude and sign of the difference: adjusted prediction value = smoothed prediction value + α * (trend line prediction value - smoothed prediction value), where α is an adjustment coefficient between 0 and 1, used to control the intensity of trend correction. The larger the α value, the closer the corrected prediction value is to the trend line, and the smaller the α value, the more characteristics of the original prediction are retained; finally, check again whether the adjusted prediction value is within the preset range.

[0175] This implementation method first uses a big data model to determine key parameters to provide a basis for correction. Then, perform smoothing according to the seasonal index and upper and lower limit values to eliminate abnormal fluctuations. Finally, through trend correction, ensure that the prediction result conforms to the long-term change trend, which can effectively improve the accuracy and reliability of water level prediction and make the prediction result more in line with the actual hydrological characteristics.

[0176] In one implementation method of this embodiment, spatially combine the corrected water level prediction value with the second three-dimensional geological model to obtain a dynamic three-dimensional geological model including water level changes in a preset future time period, including the following steps:

[0177] S701. Define the boundary and initial water level surface of the water body in the second 3D geological model;

[0178] S702. Calculate the elevation of the water level surface at each time point according to the corrected water level prediction value;

[0179] S703. Use the preset bilinear difference interpolation algorithm to spread the elevation of the water level surface at each time point to the preset 3D water level model to obtain the water level surface grid model;

[0180] S704. Integrate the water level surface grid model with the second 3D geological model to obtain the integrated model;

[0181] S705. In the integrated model, construct the 3D model at each time point, and based on the 3D model at each time point, obtain the dynamic 3D geological model including water level changes in the preset future time period.

[0182] First, define the boundary and initial water level surface of the water body in the second 3D geological model.

[0183] Specifically, first, it is necessary to determine the boundary of the water body. The water body boundary can be identified by analyzing the slope and elevation changes of the terrain. For example, the hydrological analysis tool in GIS software can be used to automatically extract the water body contour by setting the elevation threshold. Next, define the initial water level surface. The initial water level surface represents the state of the water body at the start of the simulation. It can be determined through historical observation data. The interpolation method can be used to extend the discrete water level observation points (historical observation data) to the entire water body range. In the second 3D geological model, the water body boundary and water level surface are usually represented as 3D polygons or grids. The water body boundary is a closed polygon composed of a series of 3D coordinate points (x, y, z), and the initial water level surface is a plane or surface composed of a series of 3D points with the same elevation value.

[0184] Calculating the elevation of the water level surface at each time point according to the corrected water level prediction value is a key step in converting the one-dimensional water level prediction data into a 3D spatial representation. First, determine the reference elevation of the water level prediction point. Next, convert the corrected water level prediction value into the elevation of the water level surface, that is, add the water level prediction value to the reference elevation of the water level prediction point. For example, if the predicted water level at a certain time point is 3 meters and the reference elevation of the prediction point is 50 meters, then the elevation of the water level surface at this time point is 53 meters.

[0185] Use the preset bilinear difference interpolation algorithm to spread the elevation of the water level surface at each time point to the preset 3D water level model to obtain the water level surface grid model, so as to convert the discrete water level data into a continuous surface.

[0186] Bilinear difference interpolation is a method of interpolation in two-dimensional space, which performs linear interpolation in both the x and y directions. Its advantages are fast calculation speed and the ability to generate a smooth surface.

[0187] The specific implementation steps are as follows: 1. First, create a regular grid within the water area. For example, a 10-meter * 10-meter grid can be created. 2. For each point in the grid, find the four nearest known water level points around it. These four points form a rectangle, and the point to be interpolated is located within this rectangle. 3. Use the bilinear difference interpolation formula to calculate the water level at this point. 4. Repeat step 3 until all points in the grid are assigned water level values.

[0188] The bilinear difference interpolation formula is as follows:

[0189] f(x,y)≈f(x1,y1)(x2 - x)(y2 - y)+f(z2,y1)(x - x1)(y2 - y)+f(x1,y2)(x2 - x)(y - y1)+f(x2,y2)(x - r1)(y - y1);

[0190] Where (x,y) are the coordinates of the point to be interpolated, (x1,y1), (x2,y1), (x1,y2), (x2,y2) are the coordinates of the four surrounding known points, and f(x,y) represents the function value (i.e., water level elevation) at the point (x,y).

[0191] To fuse the water level surface grid model with the second 3D geological model to obtain a fused model, the following steps are included:

[0192] 1. Align the water level surface grid model with the second 3D geological model spatially. Specifically, use the same coordinate system and spatial reference for the water level surface grid model and the 3D geological model.

[0193] 2. Convert the water level surface grid in the water level surface grid model to a format compatible with the second 3D geological model. For example, the water level surface grid can be converted to a triangular irregular network or directly inserted into the voxel model (the second 3D geological model).

[0194] 3. Through Boolean operations, connect the strata in the second 3D geological model with the water level surface in the water level surface grid model.

[0195] 4. Add the water level elevation of the water level surface grid model as attribute data to the second 3D geological model.

[0196] In the fusion model, a three-dimensional model is constructed for each time point, and based on the three-dimensional models of each time point, a dynamic three-dimensional geological model including water level changes in a preset future time period is obtained. The specific implementation process is as follows: 1. Create a separate three-dimensional model for each time point within the preset time period. One hour or one day can be selected as a time point. 2. For each time point, determine the specific water level according to the water level prediction data. 3. For each time point, update the water level information in the fusion model. Specifically, it includes updating the geometry and attribute data of the water body. 4. Link the three-dimensional models of each time point to form a continuous time series. Specifically, it can be achieved by creating a data structure containing the models of all time points, such as a dictionary or a list, where each element corresponds to the model state of a time point. In summary, a dynamic three-dimensional geological model including water level changes in a preset future time period can be obtained.

[0197] This embodiment integrates static geological information and dynamic hydrological data, can display the impact of water level changes on the geological environment in a future period of time, and helps to predict potential coal mine disasters.

[0198] In one implementation manner of this embodiment, based on the dynamic three-dimensional geological model, determining the probability of a water disaster occurring in a coal mine includes the following steps:

[0199] S801. Calculate the key parameters corresponding to the three-dimensional model of each time point. The key parameters include the volume ratio of the dangerous area, the water level rising speed, and the carrying capacity of the drainage system;

[0200] S802. Use a preset water disaster calculation formula to calculate the probability of a water disaster occurring according to the key parameters.

[0201] First, calculate the volume ratio of the dangerous area. The dangerous area is usually defined as the area where the water level is higher than a certain threshold. The volume ratio can be obtained by traversing each voxel (or grid cell) in the three-dimensional model, counting the number of voxels where the water level exceeds the safety threshold, and then dividing by the total number of voxels.

[0202] The calculation of the water level rising speed can be done by comparing the water level differences between two adjacent time points. For the entire target area, the average rising speed or the maximum rising speed can be calculated.

[0203] The carrying capacity of the drainage system is determined by the performance of the drainage pump, hydraulic losses, and hydrostatic pressure.

[0204] For example, assume that there are 3 drainage pumps in the coal mine, each with a rated flow rate of 500 cubic meters per hour and a head of 100 meters. The hydrostatic pressure is 50 meters, and the pipeline friction and local losses are calculated to be 30 meters.

[0205] The general theory of drainage capacity = 500 * 3 = 1500 cubic meters per hour, the actual working point head = 50 + 30 = 80 meters. Referring to the pump performance curve, at a head of 80 meters, the actual flow rate of each pump may drop to 450 cubic meters per hour. Then the actual total drainage capacity = 450 * 3 = 1350 cubic meters per hour. Applying a safety factor of 1.2, the carrying capacity of the drainage system = 1350 / 1.2 ≈ 1125 cubic meters per hour.

[0206] The calculation formula for the probability of water damage includes: P(water damage) = w1 * f1 (proportion of the volume of the dangerous area) + w2 * f2 (rate of water level rise) + w3 * f3 (carrying capacity of the drainage system);

[0207] Among them, w1 to w3 are the preset weights of the proportion of the volume of the dangerous area, the rate of water level rise, and the carrying capacity of the drainage system respectively, and f1 to f3 are the risk functions corresponding to the proportion of the volume of the dangerous area, the rate of water level rise, and the carrying capacity of the drainage system respectively.

[0208] In this embodiment, by calculating key parameters and applying the calculation formula for the probability of water damage, the complex three-dimensional dynamic geological model is transformed into a clear probability of water damage risk, providing an intuitive and quantitative decision-making basis for coal mine safety management, helping to identify high-risk situations in a timely manner, taking preventive measures, and effectively improving the accuracy and timeliness of water damage early warning.

[0209] The embodiment of the present application also provides a coal mine well disaster identification device based on a deep learning algorithm, including:

[0210] A memory configured to store instructions; and

[0211] A processor configured to call instructions from the memory and capable of implementing the above-mentioned coal mine well disaster identification method based on a deep learning algorithm when executing the instructions.

[0212] The embodiment of the present application also provides an electronic device, including:

[0213] The above-mentioned coal mine well disaster identification device based on a deep learning algorithm.

[0214] The embodiment of the present application also provides a machine-readable storage medium, on which instructions are stored, and the instructions are used to cause the machine to execute the above-mentioned coal mine well disaster identification method based on a deep learning algorithm.

[0215] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0216] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0217] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0218] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0219] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.

[0220] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM), and / or non-volatile memory in the form of, for example, read-only memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.

[0221] A computer-readable medium includes permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0222] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising the element.

[0223] The above are only embodiments of the present application and are not used to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A method for identifying coal mine disasters based on deep learning algorithms, characterized in that, Including: Obtain the laser scanning point cloud data, geological drilling data, and geophysical exploration data of the coal mine shaft; Divide the target area including the coal mine shaft into multiple three-dimensional grid cells, and interpolate the laser scanning point cloud data, geological drilling data, and geophysical exploration data into the corresponding three-dimensional grid cells to obtain the first three-dimensional geological model; Add stratum parameters and coal seam parameters to the first three-dimensional geological model to obtain the second three-dimensional geological model; Associate the second three-dimensional geological model with a pre-constructed big data model, obtain the current season, and obtain the monitoring data of the current season according to the big data model. The monitoring data includes weather parameters, water level data, and environmental data; Extract features from the current season, weather parameters, water level data, and environmental data to obtain a multi-dimensional time series; Input the multi-dimensional time series into a pre-constructed deep learning model to obtain the water level prediction values for each time unit in a preset future time period; Input the water level prediction values for each time unit in the preset future time period into the big data model to correct the data of the water level prediction values for each time unit in the preset future time period; Define the boundary and initial water level surface of the water body in the second three-dimensional geological model; Calculate the water level surface elevation at each time point according to the corrected water level prediction value; Use a preset bilinear difference interpolation algorithm to spread the water level surface elevation at each time point to a preset three-dimensional water level model to obtain a water level surface grid model; Perform spatial alignment between the water level surface grid model and the second three-dimensional geological model; Convert the water level surface grid in the water level surface grid model into a format compatible with the second three-dimensional geological model; Connect the strata in the second three-dimensional geological model with the water level surface in the water level surface grid model through Boolean operations; Add the water level surface elevation of the water level surface grid model as attribute data to the second three-dimensional geological model to obtain a fusion model; In the fusion model, construct a three-dimensional model for each time point, and obtain a dynamic three-dimensional geological model including water level changes in a preset future time period according to the three-dimensional model for each time point; Based on the dynamic three-dimensional geological model, determine the probability of water disaster occurrence in the coal mine shaft.

2. The method according to claim 1, wherein Divide the target area including the coal mine shaft into multiple three-dimensional grid cells, and interpolate the laser scanning point cloud data, geological drilling data, and geophysical exploration data into the corresponding three-dimensional grid cells to obtain the first three-dimensional geological model, including: Obtain the area of the target area, and determine the density of the laser scanning point cloud data according to the laser scanning point cloud data; Determine the grid resolution according to the area of the target area and the density of the laser scanning point cloud data; Divide the target area into multiple three-dimensional grid cells according to the grid resolution; In the case where any three-dimensional grid cell contains laser scanning point cloud data, geological drilling data, and / or geophysical exploration data, interpolate the laser scanning point cloud data, geological drilling data, and / or geophysical exploration data into the three-dimensional grid cell.

3. The method according to claim 2, wherein Determine the grid resolution according to the area of the target area and the density of the laser scanning point cloud data, including: Use a preset area region as a basic unit, and divide the target area into multiple sub-regions according to the basic unit; Calculate the average density and standard deviation of the laser scanning point cloud data in each sub-region; Using a preset adaptive calculation formula, calculate the grid resolution of each sub-region according to the average density and standard deviation. Among them, when the grid resolution is less than the first preset value, set the grid resolution to the first preset value, and when the grid resolution is greater than the second preset value, set the grid resolution to the second preset value, and the second preset value is greater than the first preset value.

4. The method according to claim 1, wherein Associate the second three-dimensional geological model with a pre-constructed big data model, and obtain the current season. According to the big data model, obtain the monitoring data of the current season, including: Construct an octree spatial index structure for the current season. The octree spatial index structure includes multiple cubic sub-spaces and multiple nodes storing boundary coordinates and monitoring data; Calculate the central point coordinates of each three-dimensional grid unit in the second three-dimensional geological model; According to the boundary coordinates, query the target cubic sub-space including the central point coordinates in the octree spatial index structure, and determine the node corresponding to the central point coordinates in the target cubic sub-space. The monitoring data corresponding to the central point coordinates is the monitoring data stored in the node.

5. The method according to claim 1, characterized in that The deep learning model is a long short-term memory network. The long short-term memory network includes an input layer, a first LSTM layer, a second LSTM layer, a third LSTM layer, a fully connected layer, and an output layer. Input the multi-dimensional time series into the pre-constructed deep learning model to obtain the water level prediction values of each time unit in a future preset time period, including: Input the multi-dimensional time series through the input layer, and pass through the first LSTM layer, the second LSTM layer, the third LSTM layer, and the fully connected layer respectively, and obtain the water level prediction values of each time unit in a future preset time period through the output layer.

6. The method according to claim 5, wherein Input the water level prediction values of each time unit in a future preset time period into the big data model to correct the data of the water level prediction values of each time unit in a future preset time period, including: According to the big data model, determine the seasonal index, water level upper limit value, water level lower limit value, and water level trend of the current season; Smooth the water level prediction values according to the seasonal index, water level upper limit value, and water level lower limit value; According to the water level trend, perform trend correction on the smoothed water level prediction values to obtain the water level prediction values after data correction.

7. The method according to claim 1, characterized in that, Based on the dynamic three-dimensional geological model, determine the probability of water disaster occurrence in the coal mine shaft, including: Calculate the key parameters corresponding to the three-dimensional model at each time point. The key parameters include the volume ratio of the dangerous area, the water level rising speed, and the bearing capacity of the drainage system; Adopt a preset water disaster calculation formula, and calculate the probability of water disaster occurrence according to the key parameters.

8. A coal mine disaster recognition device based on deep learning algorithm, characterized in that, Including: A memory configured to store instructions; And A processor configured to call the instructions from the memory and be able to implement the method for identifying coal mine shaft disasters based on deep learning algorithms according to any one of claims 1 to 7 when executing the instructions.

9. An electronic device, characterized in that, Including: The device according to claim 8.

Images