Artificial intelligence-based aquifer water yield property partitioning method and system

Abstract

The application relates to the technical field of geology, and provides a method and system for partitioning water-bearing layer water abundance based on artificial intelligence, which receives a multi-source heterogeneous original data set, performs data reconstruction processing on the multi-source heterogeneous original data set, and generates a reconstructed geological data body corresponding to the water-bearing layer. A water-bearing layer water abundance identification model trained in advance is called to learn the spatial features of the reconstructed geological data body, so that a preliminary water-bearing layer water abundance spatial distribution map corresponding to the water-bearing layer is obtained. The water-bearing layer is spatially divided based on the water abundance value distribution trend in the preliminary water-bearing layer water abundance spatial distribution map, a water-bearing layer water abundance partitioning and defining result corresponding to the target coal seam roof water-bearing layer is generated, and the water-bearing layer water abundance identification model and the water-bearing layer water abundance partitioning and defining result are iteratively optimized by using supplementary geological measurement data. The method can improve the prediction accuracy, prediction robustness and real-time response capability of the water-bearing layer water abundance prediction.

Description

Technical Field

[0001] This application relates to the field of geological technology, and more specifically, to an artificial intelligence-based method and system for aquifer water-bearing zoning. Background Technology

[0002] Currently, roof water hazards in coal mines are one of the core disasters threatening safe production. Accurate prediction of the water-bearing capacity of roof aquifers can effectively prevent and control these hazards. Existing technologies mainly predict the water-bearing capacity of roof aquifers using single indicators such as borehole flushing fluid consumption, unit inflow, and aquifer thickness. Alternatively, they predict the water-bearing capacity of roof aquifers by linearly superimposing geological parameter maps such as fault fractal dimension and lithological brittle-plastic index. However, existing technologies suffer from low prediction accuracy, poor robustness, and weak real-time response capabilities.

[0003] Therefore, improving the prediction accuracy, robustness, and real-time response capability of aquifer water-bearing capacity is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide an artificial intelligence-based method and system for aquifer water-bearing zoning, so as to improve the prediction accuracy, prediction robustness and real-time response capability of aquifer water-bearing.

[0005] In a first aspect, this application provides an artificial intelligence-based method for aquifer water-bearing zoning, comprising:

[0006] Receive a multi-source heterogeneous raw dataset corresponding to the aquifer on the roof of the target coal seam, wherein the multi-source heterogeneous raw dataset includes structural geological raw data, stratigraphic lithology raw data, and geophysical raw data;

[0007] The multi-source heterogeneous original dataset is reconstructed to generate a reconstructed geological data volume corresponding to the target coal seam roof aquifer. The reconstructed geological data volume includes a spatial distribution layer of structural elements, a spatial distribution layer of lithological elements, and a spatial distribution layer of geophysical attributes with spatial correlation.

[0008] The aquifer water-bearing property identification model, which has been pre-trained, is invoked to perform spatial feature learning processing on the reconstructed geological data volume. The aquifer water-bearing property identification model performs feature mapping operations at each spatial location of the reconstructed geological data volume to obtain a preliminary spatial distribution map of the aquifer water-bearing property corresponding to the target coal seam roof aquifer.

[0009] Based on the distribution trend of water-bearing value in the preliminary aquifer water-bearing spatial distribution map, the target coal seam roof aquifer is spatially divided to generate the water-bearing zoning result of the target coal seam roof aquifer.

[0010] When supplementary geological measurement data of the aquifer exposed during the mining process are obtained from the target coal seam roof aquifer, the aquifer water-bearing identification model is iteratively optimized using the supplementary geological measurement data, and the aquifer water-bearing zoning result is updated based on the optimized aquifer water-bearing identification model.

[0011] Secondly, this application provides an artificial intelligence-based aquifer water-bearing zoning system, which includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions. When the processor executes the machine-executable instructions, the artificial intelligence-based aquifer water-bearing zoning system implements the aforementioned artificial intelligence-based aquifer water-bearing zoning method.

[0012] The artificial intelligence-based aquifer water-bearing zoning method and system provided in this application receives multi-source heterogeneous raw datasets of the aquifers in the roof of the target coal seam and performs data reconstruction processing to generate a reconstructed geological data volume with spatial correlation. A pre-trained aquifer water-bearing identification model is then used to perform spatial feature learning processing on the reconstructed geological data volume to obtain a preliminary spatial distribution map of aquifer water-bearing capacity. Based on the numerical distribution trend of water-bearing capacity, spatial division is performed to generate aquifer water-bearing zoning results. Furthermore, when acquiring supplementary geological measurement data revealed during mining, this data is used to iteratively optimize the aquifer water-bearing identification model and update the zoning results. The fusion of multi-source heterogeneous data improves data comprehensiveness; the spatial feature learning ability of the artificial intelligence model characterizes the high-dimensional coupling relationship between multiple factors under complex geological conditions; and the iterative optimization mechanism enables dynamic updates based on real-time geological data. These features enhance the prediction accuracy, robustness, and real-time response capability of aquifer water-bearing capacity prediction. Attached Figure Description

[0013] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0014] Figure 1 A flowchart illustrating an artificial intelligence-based aquifer water-bearing zoning method provided in this application embodiment;

[0015] Figure 2 This is a schematic diagram of the structure of an aquifer water-bearing zoning system based on artificial intelligence, provided in an embodiment of this application.

[0016] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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.

[0018] Figure 1 This is a flowchart illustrating an artificial intelligence-based aquifer water-bearing zoning method provided in this application embodiment. It should be understood that in other embodiments, the order of some steps in the artificial intelligence-based aquifer water-bearing zoning method of this embodiment can be shared according to actual needs, or some steps can be omitted or maintained. Figure 1 As shown, the method may include the following steps:

[0019] S110. Receive the multi-source heterogeneous raw dataset corresponding to the aquifer on the roof of the target coal seam. The multi-source heterogeneous raw dataset includes structural geological raw data, stratigraphic lithology raw data, and geophysical raw data.

[0020] The aquifer above the target coal seam refers to the water-bearing rock strata located above the target coal seam and having a direct or indirect hydraulic connection with the coal seam. Multi-source heterogeneous raw datasets refer to data sets originating from different exploration methods, with different data formats and storage structures. These data collectively describe the structural characteristics, lithological characteristics, and geophysical properties of the aquifer above the target coal seam.

[0021] Raw structural geological data can include, for example, fault feature data, fold feature data, and joint and fracture development data. Fault feature data can include information such as fault strike, dip, dip angle, fault displacement, and fracture zone width. Fold feature data can include information such as the axial position and limb dip angle of anticlines and synclines. Joint and fracture data can include information such as joint density, joint attitude, and fracture opening width.

[0022] Raw stratigraphic lithological data may include, for example, borehole core logging data, lithological thickness data, and rock physical and mechanical property test data. Borehole core logging data may include borehole opening coordinates, borehole depth, and lithological names for different depth ranges. Lithological thickness data may include the cumulative thickness of each lithological layer in a single borehole. Rock physical and mechanical property test data may include parameters such as uniaxial compressive strength, tensile strength, elastic modulus, and Poisson's ratio.

[0023] The raw geophysical data can specifically include 3D seismic exploration data and transient electromagnetic exploration data. 3D seismic exploration data can include raw acquired seismic data, processed stacked data volumes, migration data volumes, and seismic attribute data such as root-mean-square amplitude and instantaneous frequency extracted along the target horizon. Transient electromagnetic exploration data can include field-measured secondary field attenuated voltage data and resistivity data obtained through one-dimensional or two-dimensional inversion calculations.

[0024] When receiving multi-source heterogeneous raw datasets, for structural geology and stratigraphic lithology data stored in relational databases, structured query language statements can be executed through the database connection interface to extract the required fields and data records from the corresponding data tables. For geophysical data files stored in standard formats, the corresponding data reading library can be called to parse the file header information and data body according to the standard format. During the data reception process, if the raw data contains sensitive spatial location information such as the precise coordinates of boreholes, anonymization processing can be performed using coordinate translation.

[0025] S120. Perform data reconstruction processing on the multi-source heterogeneous original dataset to generate a reconstructed geological data volume corresponding to the target coal seam roof aquifer. The reconstructed geological data volume includes a spatial distribution layer of structural elements, a spatial distribution layer of lithological elements, and a spatial distribution layer of geophysical attributes with spatial correlation.

[0026] The purpose of data reconstruction processing is to eliminate the multi-source heterogeneity of the original data, transform it into a unified coordinate system, unified raster unit, and unified attribute dimension, and generate a standardized data volume in which each spatial location point is associated with a complete set of geological information.

[0027] A structural feature spatial distribution layer refers to a set of continuously distributed raster layers converted from discrete structural geological raw data using spatial interpolation methods. When generating a structural feature spatial distribution layer, all point or line data related to structural geology are first extracted from a multi-source heterogeneous raw dataset and converted into point data suitable for spatial interpolation.

[0028] For fault distribution data, multiple sampling points can be generated along the fault line at certain intervals based on the spatial location and displacement of each fault. Each sampling point carries the fault displacement value at that location. For joint fracture data, point data can be generated based on the location of joint measurement points. Each point carries the joint density value at that measurement point. These points with spatial coordinates and attribute values ​​constitute the input sample point set for spatial interpolation.

[0029] For example, Kriging interpolation can be used to spatially interpolate these discrete points. For instance, spatial variability analysis can be performed on the sample point data to calculate the experimental variability function. For a given lag distance h, all pairs of sample points at distance h are counted, the square of the difference in attribute values ​​between each pair is calculated, and then half of the average is taken to obtain the experimental variability function value corresponding to that lag distance h. Repeating the above calculation for different lag distances yields a series of experimental variability function values.

[0030] Next, a theoretical variogram model can be selected to fit the experimental variogram values. For example, the least squares method can be used to find the model parameters that minimize the sum of squared residuals between the theoretical and experimental values. After fitting the theoretical variogram model, for the grid point to be interpolated, known sample points within a certain range around it are found. Based on the spatial relationships between these sample points and their spatial relationships with the point to be interpolated, a system of Kriging equations is constructed. Then, the weight coefficients of each known sample point are obtained by solving this system of linear equations. Finally, the attribute estimate of the point to be interpolated is calculated as a weighted sum of the attribute values ​​of all known sample points involved in the estimation and their weight coefficients.

[0031] By performing the above Kriging interpolation calculation on all grid points within the entire study area, continuous attribute value raster maps can be obtained, such as fault displacement distribution raster maps and joint density distribution raster maps. These raster layers together constitute the spatial distribution layer of structural elements.

[0032] The spatial distribution layer of lithological elements refers to a set of raster layers that reflect the spatial variation of lithological combination characteristics, generated from borehole lithological data through three-dimensional geological modeling and attribute extraction methods.

[0033] When generating a spatial distribution layer of lithological elements, lithological information at different depths for each borehole can be extracted from borehole core logging data. This includes borehole opening coordinates and the top and bottom boundary depths and lithological names of each lithological stratum. Using this borehole data to construct a three-dimensional lithological model of the target coal seam's roof aquifer, the target stratum can be determined first based on the research objective. For each lithological type, the borehole data is converted into indicator variables. At each depth point in each borehole, if the lithology at that point matches the target lithology, the indicator variable is assigned a value of 1; otherwise, it is assigned a value of 0. Then, three-dimensional spatial interpolation is performed on the indicator variables for each lithology type to generate a three-dimensional indicator probability volume. The value of each volume element represents the probability that the location belongs to that lithology. For each three-dimensional volume element, by comparing the probabilities of belonging to different lithologies, the lithology type with the highest probability is used as the final lithological assignment for that volume element, thus obtaining a complete three-dimensional lithological model.

[0034] When extracting a two-dimensional raster layer reflecting lithological assemblage characteristics based on a three-dimensional lithology model, the cumulative thickness of sandstone within the three-dimensional column corresponding to each planar grid position can be calculated by summing the thicknesses of all sandstone-type volume elements below that grid position. Similarly, the cumulative thickness of mudstone can be calculated, and the cumulative thickness of sandstone can be divided by the cumulative thickness of mudstone to obtain the sand-mud ratio value for that grid point. This calculation is then performed on all grid points to generate a raster layer showing the spatial distribution of the sand-mud ratio.

[0035] In addition, lithological diversity indices can be calculated, such as the Shannon diversity index. First, the volume proportion of each major lithological type below each grid location is calculated. Then, the negative proportion of each lithology is multiplied by the sum of the natural logarithms of that proportion. This calculation is performed on all grid points to generate a spatial distribution raster layer of lithological diversity indices. These two-dimensional raster layers derived from the three-dimensional lithological model together constitute the spatial distribution layer of lithological elements.

[0036] A geophysical attribute spatial distribution layer refers to a set of raster layers reflecting the spatial variations of geophysical field characteristics, generated through processing, inversion, and attribute extraction of raw geophysical data. When generating a geophysical attribute spatial distribution layer for 3D seismic exploration data, the raw seismic data can first be preprocessed to improve the signal-to-noise ratio and resolution. Then, velocity analysis is performed to establish a velocity field, and migration processing is carried out on this basis to obtain a 3D seismic data volume that accurately reflects the subsurface structural morphology. Next, the seismic reflection horizons of the target coal seam roof are interpreted and tracked. After obtaining continuous roof horizon interpretation data, a fixed time window is opened upwards or downwards along this horizon, and various seismic attributes within the time window are extracted. When extracting the root mean square amplitude attribute, for each seismic trace, the amplitude values ​​of all sampling points can be extracted within a specified time window. The sum of the squares of these amplitude values ​​is calculated, divided by the number of sampling points, and then the square root is taken. The result is used as the root mean square amplitude value of the seismic trace. This value is assigned to the ground coordinate point corresponding to the seismic trace to generate an amplitude attribute point set. This operation is repeated for all seismic traces, and the point set is meshed and interpolated to generate a root mean square amplitude attribute raster layer.

[0037] When extracting instantaneous frequency attributes, a Hilbert transform can be performed on the original seismic trace to obtain a complex seismic trace. The instantaneous phase of the complex seismic trace can be calculated, and the instantaneous frequency is the derivative of the instantaneous phase with respect to time. The average value of the instantaneous frequencies of all sampling points within the target time window is taken as the representative value of the seismic trace, thereby generating an instantaneous frequency attribute raster layer.

[0038] For transient electromagnetic exploration data, the initial observation data, i.e., the secondary field voltage values ​​measured at different delay times, can be read first to establish an initial model of the subsurface medium, such as a one-dimensional layered model. Based on the initial model, the theoretical induced voltage decay curve is calculated, and the model parameters are continuously adjusted through an inversion algorithm to minimize the fitting difference between the theoretical decay curve and the measured decay curve.

[0039] After successful inversion, resistivity values ​​at various depths below each measuring point are obtained. For the target depth range where the aquifer on the roof of the target coal seam is located, the resistivity values ​​within that depth range are extracted, and their arithmetic mean is calculated as the representative resistivity of the target layer at that measuring point. The representative resistivity values ​​of all measuring points are then interpolated using a grid based on their spatial coordinates to generate an apparent resistivity attribute raster layer. These raster layers derived from geophysical data collectively constitute the spatial distribution layer of geophysical attributes.

[0040] After generating the spatial distribution layers of structural elements, lithological elements, and geophysical attributes, these layers can be spatially registered and overlaid to generate the final reconstructed geological data volume.

[0041] Spatial registration refers to converting all layers to the same spatial reference system. If the projection methods or coordinate systems of different layers are inconsistent, projection transformation is required. Spatial registration also includes unifying the cell size of all layers, i.e., resampling. Overlay refers to combining multiple single-band raster layers into a multi-band raster data volume. For each spatial coordinate point in the study area, the fault density value of that point is read from the spatial distribution layer of structural elements, the sand-mud ratio value of that point is read from the spatial distribution layer of lithological elements, and the root mean square amplitude attribute value of that point is read from the spatial distribution layer of geophysical attributes. These values ​​together constitute a multi-dimensional feature vector describing the geological and geophysical characteristics of that spatial location. The set of these multi-dimensional feature vectors of all spatial locations is the reconstructed geological data volume.

[0042] S130. Call the aquifer water-bearing identification model that has completed the training process in advance to perform spatial feature learning processing on the reconstructed geological data volume. Through the aquifer water-bearing identification model, perform feature mapping operation at each spatial location of the reconstructed geological data volume to obtain a preliminary aquifer water-bearing spatial distribution map corresponding to the aquifer on the roof of the target coal seam.

[0043] The aquifer water-bearing capacity identification model is a predictive model built on machine learning algorithms. Its core function is to learn and reconstruct the complex nonlinear mapping relationship between the multi-source geological elements contained in the geological data volume and the actual water-bearing capacity of the aquifer.

[0044] The aquifer water-bearing capacity identification model has been trained offline using measured water-bearing capacity data from historically existing exploration boreholes and pumping test wells, along with the corresponding features of the reconstructed geological data volume. The input to the aquifer water-bearing capacity identification model is a multi-dimensional feature vector, i.e., the feature vector of a certain spatial location point in the reconstructed geological data volume. The output of the aquifer water-bearing capacity identification model is a continuous water-bearing capacity prediction value.

[0045] Calling the aquifer water-bearing identification model to perform spatial feature learning processing on the reconstructed geological data volume essentially involves having the trained aquifer water-bearing identification model perform a forward calculation on every spatial location point within the entire study area, thereby generating a preliminary spatial distribution map of aquifer water-bearing properties that covers the entire study area.

[0046] S131. The reconstructed geological data volume is segmented according to spatial coordinate points to obtain multiple spatial point feature vectors. Each spatial point feature vector contains the values ​​of structural elements, lithological elements, and geophysical attributes at the corresponding coordinate point.

[0047] The reconstructed geological data volume can be viewed as a regular three-dimensional grid structure, with two dimensions being spatial coordinates X and Y, and a third dimension being the feature dimension. Processing the reconstructed geological data volume can begin by segmenting it according to each spatial coordinate point. Assume the spatial range covered by the reconstructed geological data volume extends from the starting value of the X coordinate to the ending value of the X coordinate, and from the starting value of the Y coordinate to the ending value of the Y coordinate, with a set spatial resolution. Multiple grid points can be divided along the X direction, and multiple grid points can be divided along the Y direction, resulting in a total of multiple spatial grid points within the entire study area.

[0048] For grid points with specific indices, corresponding values ​​can be extracted from various bands of the reconstructed geological data volume. Fault density values ​​are extracted from the bands corresponding to the spatial distribution layer of structural features; sand-to-mud ratio values ​​are extracted from the bands corresponding to the spatial distribution layer of lithological features; and root-mean-square amplitude and apparent resistivity values ​​are extracted from the bands corresponding to the spatial distribution layer of geophysical attributes. These extracted values, arranged in a predetermined order, are organized into a one-dimensional array to obtain the spatial feature vector corresponding to the spatial coordinates. By traversing all grid points, all spatial feature vectors are obtained.

[0049] S132. Input the spatial point feature vector into the aquifer water-bearing property identification model, determine the final leaf node to which the spatial point feature vector belongs in the decision tree group structure of the aquifer water-bearing property identification model, and the water-bearing property reference value stored in the final leaf node, and obtain a preliminary discrimination result.

[0050] The feature vector of each spatial point is input into the aquifer flooding identification model. The aquifer flooding identification model is internally a decision tree group structure composed of multiple decision trees. For each decision tree, the input feature vector will undergo a path selection process from the root node to the leaf node and finally fall on a certain leaf node.

[0051] S1321. Input the spatial point feature vectors into each decision tree in the decision tree group structure, and perform layer-by-layer matching processing on each decision tree starting from the root node.

[0052] For the k-th decision tree in a group of decision trees, the spatial point feature vector can be input into the root node of the tree. The root node is the starting node of the decision tree and stores the splitting rules for partitioning the data. These rules include a splitting feature dimension identifier to indicate which dimension's value should be selected from the input feature vector for comparison, and a splitting threshold to serve as the benchmark for comparison.

[0053] S1322. Extract the corresponding feature value from the spatial point feature vector according to the split feature dimension identifier stored in the current decision tree node, and compare the feature value with the split threshold stored in the current decision tree node to determine the next branch child node.

[0054] At the root node, based on the split feature dimension identifier stored in the root node, the specific value of that dimension is extracted from the spatial point feature vector. Then, this value is compared with the split threshold stored in the root node. If the value is less than or equal to the split threshold, the current feature vector is assigned to the left child node of the root node; if the value is greater than the split threshold, it is assigned to the right child node of the root node.

[0055] S1323. After updating the next branch sub-node to a new current decision tree node, repeat the layer-by-layer matching process until the leaf node of the decision tree is reached, and then determine the leaf node as the final leaf node to which the spatial point feature vector belongs on the decision tree.

[0056] After entering the left or right child node, the child node is updated as the new current decision tree node. This new current node also stores its own split feature dimension identifier and split threshold. Then, the value of the corresponding dimension is extracted from the feature vector and compared with the split threshold of the node to determine the branch child node of the next layer.

[0057] This process is repeated. Each time a new node is reached, the corresponding dimension value is extracted from the feature vector according to the splitting rules stored in that node. The value is then compared and the node moves down along the branch that meets the conditions. This process continues until the bottom node of the decision tree, i.e., the leaf node, is reached.

[0058] When a leaf node is reached, the traversal of the feature vector of that spatial point on the decision tree ends, and that leaf node is the final leaf node to which it belongs on the decision tree.

[0059] S1324. Read the average value of the water abundance of the samples in the training sample set covered by the final assigned leaf node from the data stored in the final assigned leaf node, and use the average value as the preliminary discrimination result of the current decision tree outputting the feature vector of the spatial point.

[0060] During the model training phase, each leaf node stores statistical information about the true water abundance values ​​of all training samples falling into that node. For the final leaf node of the k-th decision tree, this leaf node stores the average of the true water abundance values ​​of all sample points falling into this leaf node from the training set during model training.

[0061] When the feature vector of a spatial point reaches the leaf node of the k-th decision tree, the stored average value is read from that node and used as the preliminary discrimination result of the k-th decision tree on the feature vector of the spatial point. The above steps are repeated for each decision tree in the decision tree group structure to obtain the preliminary discrimination results of all decision trees.

[0062] S133. The preliminary discrimination results output by each decision tree in the aquifer water-bearing identification model are aggregated, and the statistical average of the preliminary discrimination results output by all decision trees is taken as the water-bearing discrimination output value corresponding to the spatial point feature vector.

[0063] After obtaining the preliminary discrimination results of all decision trees on the feature vectors of spatial points, these preliminary discrimination results are aggregated to obtain the final comprehensive prediction value. For example, aggregation can be performed by averaging: the preliminary discrimination results output by all decision trees are added together to obtain a sum, and then this sum is divided by the total number of decision trees to obtain the average value. This average value is the final output value of the aquifer water-bearing identification model for the water-bearing prediction at the current spatial coordinate point, that is, the water-bearing discrimination output value corresponding to the feature vector of that spatial point.

[0064] S134. The water-bearing discrimination output value is backfilled according to the spatial coordinate points corresponding to the water-bearing discrimination output value to generate a two-dimensional water-bearing numerical distribution matrix covering the reconstructed geological data volume space.

[0065] After calculating the water-bearing capacity output value for each spatial coordinate point, these values ​​are reorganized into a two-dimensional matrix that matches the spatial extent of the original reconstructed geological data volume. The row index of this two-dimensional matrix corresponds to the grid point number in the Y direction, and the column index corresponds to the grid point number in the X direction.

[0066] For a grid with a corresponding number of rows and columns, a two-dimensional matrix of the same size is created. The element in the i-th row and j-th column of the matrix is ​​assigned the water-bearing discrimination output value of the corresponding spatial coordinate point. After traversing all grid points and assigning values ​​to all matrix elements, a complete two-dimensional water-bearing numerical distribution matrix is ​​generated. This two-dimensional water-bearing numerical distribution matrix accurately records the model-predicted water-bearing values ​​at each spatial location point within the study area.

[0067] S135. The two-dimensional water-bearing numerical distribution matrix is ​​converted into a visualized raster image to obtain the preliminary spatial distribution map of water-bearing properties of the aquifer.

[0068] The two-dimensional water-rich numerical distribution matrix generated in step S134 can be converted into a visualized raster image using geographic information system software.

[0069] Specifically, a two-dimensional water-bearing capacity distribution matrix can be used as an array of cell values ​​for raster data. Spatial reference information, including geographic extent, cell size, and coordinate system, is defined for this raster data. This spatial reference information is then written into a standard raster image file format, where each cell value in the generated file represents the predicted water-bearing capacity value at that cell's location.

[0070] Optionally, to facilitate human interpretation, different colors can be assigned to different ranges of water-bearing values, thereby generating an intuitive preliminary spatial distribution map of aquifer water-bearing properties.

[0071] S140. Based on the distribution trend of water-bearing values ​​in the preliminary aquifer water-bearing spatial distribution map, the aquifer on the roof of the target coal seam is spatially divided to generate the water-bearing zoning result of the aquifer corresponding to the aquifer on the roof of the target coal seam.

[0072] After obtaining a preliminary spatial distribution map of the aquifer's water-bearing properties, the roof aquifer of the target coal seam can be spatially divided based on the distribution trend of the water-bearing values ​​in this preliminary spatial distribution map, in order to generate water-bearing zoning results with clear boundaries. This process aims to divide a continuous water-bearing value distribution map into several relatively homogeneous internal blocks with significant differences between them, according to certain rules.

[0073] S141. Extract the water-bearing values ​​of all spatial locations from the preliminary aquifer water-bearing spatial distribution map, and arrange the extracted water-bearing values ​​in ascending order to generate a water-bearing value sequence.

[0074] The water-bearing values ​​recorded in all pixels within the preliminary aquifer water-bearing spatial distribution map generated in step S135 are extracted. This preliminary aquifer water-bearing spatial distribution map contains multiple pixels, each corresponding to a water-bearing value. All these water-bearing values ​​are extracted to form a set of water-bearing values.

[0075] Then, all the values ​​in the water-rich data set can be sorted. A sorting algorithm can be used to arrange them in ascending order to generate an ordered sequence of water-rich data.

[0076] S142. Calculate the degree of difference between adjacent water-rich values ​​in the water-rich value sequence, and select multiple water-rich values ​​from the water-rich value sequence based on the change characteristics of the degree of difference to generate a candidate set of grading boundaries.

[0077] The degree of difference here can be represented by the difference between the previous and subsequent values. For example, the difference between the t-th element and the (t+1)-th element in the sequence can be calculated to obtain a difference sequence containing multiple differences. These differences reflect the gradient of change in water abundance values ​​after sorting. Larger differences indicate a jump or discontinuity in the distribution of water abundance values, which may indicate the natural boundary between different water abundance levels.

[0078] Then, based on the variation characteristics of these differences, possible grading boundaries are screened. The mean and standard deviation of all differences are calculated, and the larger of two adjacent values ​​corresponding to differences greater than a set threshold is marked as a potential candidate grading boundary value. The set threshold can be, for example, the mean plus a set factor multiplied by the standard deviation.

[0079] In this way, a series of water-bearing values ​​can be selected from the entire water-bearing value sequence as a candidate set of classification boundaries.

[0080] S143. Determine the target grading boundary from the grading boundary candidate set based on the numerical intervals corresponding to the grading boundary candidate values, the water-bearing numerical discrete characteristics within each numerical interval, and the water-bearing numerical difference characteristics between each numerical interval.

[0081] After obtaining the candidate set of classification boundaries, the most reasonable target classification boundaries can be finally determined from them. For example, assuming the plan is to classify the aquifer into three levels, two target classification boundaries need to be selected from the candidate set.

[0082] Specifically, the selection principle is to comprehensively consider maximizing the difference between intervals and minimizing the difference within each interval. For example, the idea of ​​the natural breakpoint method can be used for optimization selection, selecting P target classification boundaries from the candidate set. The number of P is equal to the planned number of levels minus 1. These boundaries divide the entire water-bearing numerical range into P plus 1 numerical intervals.

[0083] For each possible combination of boundaries, the overall classification appropriateness can be calculated based on the sum of squared deviations within each sub-interval and the overall sum of squared deviations. For example, one can first calculate the mean of all water-bearing values ​​within each sub-interval, and then calculate the sum of squared deviations of each value within that sub-interval from the mean, i.e., the variance contribution within that sub-interval. The sum of squared deviations within all sub-intervals is then added together to obtain the sum of squared deviations within the overall category. Simultaneously, the global mean of all water-bearing values ​​within the entire study area and the overall sum of squared deviations of all values ​​relative to this global mean are calculated to minimize the sum of squared deviations within the overall category, i.e., to maximize the homogeneity within each sub-interval.

[0084] S144. Based on the target classification boundary, determine the water-bearing level of each spatial location point in the preliminary aquifer water-bearing spatial distribution map, and assign the corresponding water-bearing level label to the spatial location points whose water-bearing values ​​are in different value ranges.

[0085] Once the target classification boundaries are determined, the water abundance level of all spatial locations can be assigned.

[0086] Specifically, the entire range of water-bearing values ​​can be divided into multiple intervals based on the defined target classification boundaries, and water-bearing level labels can be set for these intervals. Each cell in the preliminary aquifer water-bearing spatial distribution map is traversed, its water-bearing value is read, and a corresponding level label is assigned to it according to the value interval in which the value belongs.

[0087] S145. Perform spatial adjacency analysis on spatial location points with the same water abundance level identifier, and group spatial location points that are adjacent in spatial location and have the same water abundance level identifier into the same water abundance level block.

[0088] In this step, after assigning a level label to each spatial location point, these labels will exhibit a patchy distribution in space. Connectivity analysis methods from image processing can be used to aggregate discrete points into continuous planar regions. For example, the eight-neighborhood algorithm can be used to start with a cell with a specific level label and examine the eight neighboring cells in its eight directions. If these neighboring cells also have the same level label, they can be classified as the same connected component.

[0089] Then, the search continues outward from the newly assigned cell until no more adjacent cells with the same grade label can be found. All spatially connected cells with the same grade label formed a connected region, i.e., a water-rich grade block. The same connectivity analysis is performed on all grade labels, and the entire study area is divided into several spatially continuous blocks with consistent internal water-rich grade labels.

[0090] S146. Generate block boundary geometric information for each of the water-bearing grade blocks, and associate the block boundary geometric information with the corresponding water-bearing grade identifier and store it as the water-bearing zoning result of the aquifer.

[0091] In this step, boundary geometry information can be generated for each water-rich block obtained through connected component analysis. Specifically, for a block consisting of multiple consecutive pixels, a boundary tracing algorithm can be used to extract its outer contour.

[0092] Specifically, a raster vectorization method can be used to convert the set of all cells constituting the block into polygon vector features. A boundary tracing algorithm starts from a boundary cell and searches along the edge of the block, recording a series of coordinate points that form the boundary, ultimately creating a closed loop. If a block contains cavities (i.e., is surrounded by blocks of other levels), the geometric information of both the outer and inner boundaries is generated simultaneously. The generated polygon vector features are associated with and stored with the corresponding water-bearing level identifier. A layer is created in a geospatial database containing fields storing the polygon vector boundaries of each block and fields storing the block's level identifier, thereby generating the water-bearing zoning results corresponding to the aquifer in the target coal seam roof.

[0093] S150. When the supplementary geological measurement data of the aquifer on the roof of the target coal seam is obtained during the mining process, the aquifer water-bearing identification model is iteratively optimized using the supplementary geological measurement data, and the aquifer water-bearing zoning result is updated based on the optimized aquifer water-bearing identification model.

[0094] As mining operations progress, when tunnel excavation or face mining exposes the aquifer aquifer on the roof of the target coal seam, new and more direct geological and hydrogeological data can be obtained. This supplementary geological data powerfully complements and validates existing exploration data. Using this newly acquired data, the previously used aquifer water-bearing identification model can be iteratively optimized to improve its predictive accuracy, thereby updating the original water-bearing zoning results.

[0095] S151. From the supplementary geological measurement data, the newly revealed spatial location point, the newly revealed structural geological data, the newly revealed stratigraphic lithology data, the newly revealed geophysical data, and the newly revealed measured water-bearing data are extracted.

[0096] During underground mining, when a roadway is exposed to a specific location, the following information can be extracted from the supplementary geological measurement data at that point: the newly exposed spatial location point, i.e., the three-dimensional coordinates of that point; newly exposed structural geological data, such as the fault displacement of small faults observed at that point and the width of the fault influence zone; newly exposed lithological data, such as the lithological name of the roof at that point and the thickness of the strata; newly exposed geophysical data, which may include local geophysical attribute values ​​obtained from underground geophysical exploration at that point; and newly exposed measured water-bearing data, such as the unit water inflow measured by drilling water exploration boreholes at the exposed point and conducting water release tests.

[0097] S152. Perform data reconstruction processing on the newly revealed structural geological data, the newly revealed stratigraphic lithology data, and the newly revealed geophysical data to generate a new revealed reconstructed geological data body with the same data structure as the reconstructed geological data body.

[0098] In order to input the newly revealed data into the aquifer water-bearing identification model, the data needs to be processed to have the same data structure as the reconstructed geological data volume generated in step S120.

[0099] Based on the spatial coordinates of the newly revealed point, a spatial feature vector corresponding to that point can be generated. Each dimension of this spatial feature vector corresponds to the feature dimensions used during model training. Then, the values ​​of these feature dimensions can be calculated based on the newly revealed raw data.

[0100] Specifically, for features such as fault density, the value of the newly revealed point can be calculated using the same interpolation method as before, based on fault information observed near the point and the tectonic background of the area where the point is located, or the value can be directly extracted from the existing spatial distribution layer of tectonic elements by spatial location. For features such as sand-mud ratio, the representative value of the point can be calculated by combining the lithology revealed by the newly revealed point and the lithology information of surrounding boreholes. For geophysical properties, the value of the point can be directly extracted from the existing spatial distribution layer of geophysical properties by spatial location.

[0101] Finally, the calculated or extracted values ​​are organized in the same order as before to form a new revealed and reconstructed geological data volume, i.e., the spatial point feature vector corresponding to the point, which is consistent with the model input format.

[0102] S153. Input the newly revealed reconstructed geological data into the aquifer water-bearing identification model for water-bearing prediction processing to obtain the predicted water-bearing value corresponding to the newly revealed spatial location point. Determine the model deviation of the aquifer water-bearing identification model based on the deviation between the predicted water-bearing value and the newly revealed measured water-bearing data.

[0103] The newly generated reconstructed geological data volume is input into the current aquifer water-bearing identification model, and the same prediction process as before is executed. The aquifer water-bearing identification model can process the new input data based on its current decision tree group structure, and finally output the predicted water-bearing value.

[0104] Then, the predicted water-bearing value of the aquifer water-bearing identification model is compared with the water-bearing value obtained by downhole measurement at that point. The deviation between the two is calculated in the form of absolute error or squared error. This deviation reflects the prediction accuracy of the current aquifer water-bearing identification model at the newly revealed point and can be defined as the model deviation.

[0105] S154. Based on the model deviation, the decision tree group structure inside the aquifer water-bearing identification model is corrected, the water-bearing reference data stored in the leaf nodes of the decision tree group structure is updated, and the feature selection criteria used for path division in the decision tree nodes are adjusted so that the difference between the predicted water-bearing value output by the corrected aquifer water-bearing identification model at the newly revealed spatial location point and the newly revealed measured water-bearing data is less than a preset difference threshold, thereby obtaining the optimized aquifer water-bearing identification model.

[0106] When the model bias is large, such as exceeding a preset threshold, the current aquifer flooding identification model needs to be corrected. The goal of this correction is to adjust the internal parameters of the aquifer flooding identification model so that its predictions on new data points more closely match the measured results. Specifically, the correction process may include updating the stored values ​​of relevant leaf nodes in the decision tree group structure, and fine-tuning the decision tree node splitting rules.

[0107] S155. The reconstructed geological data volume is re-inputted into the optimized aquifer water-bearing identification model for water-bearing prediction processing to obtain an updated preliminary aquifer water-bearing spatial distribution map.

[0108] After refining the model to obtain the optimized aquifer water-bearing identification model, the original reconstructed geological data volume covering the entire study area needs to be input back into this optimized model. Since the decision tree group structure within the model has been adjusted based on the newly revealed measured data, its parameters have changed, and therefore the model's output for the same input data will also change accordingly.

[0109] The purpose of re-inputting the reconstructed geological data volume into the model is to use this optimized model to re-predict the water-bearing capacity of every spatial point within the entire study area, thereby generating a more accurate spatial distribution map of water-bearing capacity. This newly generated spatial distribution map will serve as the basis for subsequent updates to zoning delineation.

[0110] The reconstructed geological data volume is a multi-band raster data volume that contains the values ​​of structural elements, lithological elements, and geophysical attributes at all spatial coordinate points within the study area.

[0111] The reconstructed geological data volume is then input into the optimized model, and predictions are made point by point in space, following the same processing logic as steps S131 to S134.

[0112] Specifically, the reconstructed geological data volume can first be segmented according to spatial coordinate points to obtain spatial feature vectors corresponding to each grid point. For a grid point with a specific index, its feature vector consists of the fault density value, sand-to-mud ratio value, root-mean-square amplitude attribute value, and apparent resistivity value for that point. The fault density value reflects the intensity of the influence of fault structures on that point, the sand-to-mud ratio value reflects the grain size characteristics of the lithological assemblage at that point, the root-mean-square amplitude attribute value reflects the attenuation characteristics of seismic wave energy at that point, and the apparent resistivity value reflects the electrical characteristics of the strata at that point. These values ​​extracted from different layers correspond precisely in space and together constitute a comprehensive vector describing the geological and geophysical characteristics of that point.

[0113] The aforementioned spatial point feature vectors are sequentially input into the optimized aquifer water-bearing capacity identification model. For each input feature vector, the optimized model performs a complete forward computation process, which is the same as the model computation process described above and will not be repeated here.

[0114] In the optimized aquifer flooding identification model, the average values ​​read may differ from those in the original model because the stored values ​​of the relevant leaf nodes were adjusted based on newly revealed data. By performing the aforementioned traversal and reading operations on each decision tree within the aquifer flooding identification model, a series of preliminary discrimination results equal to the number of decision trees can be obtained.

[0115] After obtaining the preliminary discrimination results of all decision trees, the optimized aquifer water-bearing identification model aggregates these results to obtain the final water-bearing discrimination output value of the optimized aquifer water-bearing identification model for the current input spatial point feature vector.

[0116] After the water-bearing discrimination output values ​​for all spatial points have been calculated according to the above process, these water-bearing discrimination output values ​​are backfilled according to their corresponding spatial coordinates. That is, for grid points with specific coordinates, the calculated water-bearing discrimination output value is assigned to the element at the corresponding position in a two-dimensional matrix. By traversing all grid points and completing the assignment of values ​​to all matrix elements, a complete two-dimensional water-bearing numerical distribution matrix covering the entire study area is generated. Each row and each column of this two-dimensional water-bearing numerical distribution matrix corresponds to a specific location on the ground, and each value in the two-dimensional water-bearing numerical distribution matrix is ​​the updated model-predicted water-bearing value for that location.

[0117] Finally, the aforementioned two-dimensional matrix is ​​converted into a visualized raster image, resulting in an updated preliminary spatial distribution map of aquifer water-bearing capacity. Compared to the initially generated preliminary distribution map, this updated map corrects the predicted values ​​for points with newly revealed measured data and their surrounding areas, thus enabling the entire distribution map to more accurately reflect the actual underground conditions.

[0118] S156. Perform spatial division processing based on the updated preliminary aquifer water-bearing spatial distribution map to obtain the updated aquifer water-bearing zoning results.

[0119] After obtaining the updated preliminary spatial distribution map of aquifer water-bearing capacity, the spatial partitioning process can be re-executed based on this new distribution map to generate updated aquifer water-bearing capacity zoning results.

[0120] Since the water-bearing values ​​on the updated distribution map have changed, especially in areas affected by newly revealed data, their distribution trends may differ from the original distribution map. Therefore, the original zoning boundaries may no longer be applicable and need to be redefined based on the new data. The specific steps for performing spatial zoning are the same as steps S140 and its sub-steps S141 to S146, but the object of the processing changes from the original preliminary aquifer water-bearing spatial distribution map to the updated distribution map.

[0121] Specifically, the water-bearing values ​​of all pixels can be extracted from the updated preliminary aquifer water-bearing spatial distribution map to form a numerical set. This numerical set contains the predicted values ​​for all points within the updated study area. Then, all values ​​in this numerical set are sorted using algorithms such as quicksort or mergesort, arranging them in ascending order to generate a new water-bearing value sequence. The purpose of sorting is to observe the distribution patterns of the values ​​and to prepare for the subsequent search for natural discontinuities.

[0122] Next, the degree of difference between adjacent values ​​in this new numerical sequence is calculated. For each position in the sequence, the difference between the next value and the previous value is calculated, resulting in a difference sequence. These differences reflect the gradient of change in water abundance values ​​after sorting. Larger differences indicate a jump or discontinuity in the distribution of water abundance values, which may indicate natural boundaries between different water abundance levels. The variation characteristics of this difference sequence are analyzed, and the mean and standard deviation of all differences are calculated. A screening threshold is set, for example, the mean plus a certain factor multiplied by the standard deviation. The larger of the two adjacent values ​​corresponding to differences greater than this threshold is marked as a potential candidate value for the classification boundary.

[0123] In this way, a series of water-bearing values ​​can be selected from the entire water-bearing value sequence to form a new candidate set of grading boundaries.

[0124] Next, an optimization process using the natural breakpoint method is employed to select a set of optimal target classification boundaries from the new candidate set. For example, assuming the plan is to classify the aquifer into three levels, two target classification boundaries need to be selected from the candidate set. The selection process requires traversing all possible combinations of boundaries.

[0125] For each possible combination, the entire range of water abundance values ​​needs to be divided into three intervals based on these two boundaries: a weakly water-rich interval, a moderately water-rich interval, and a strongly water-rich interval. Then, the classification effect under this division method is calculated.

[0126] The classification performance can be measured using a classification fitness index. For example, first, calculate the sum of squared deviations of all water-bearing values ​​within each interval from the mean of that interval. Add the sums of squared deviations of the three intervals to obtain the overall sum of squared deviations within each category. Simultaneously, calculate the global mean of all water-bearing values ​​within the entire study area, and the total sum of squared deviations of all values ​​relative to this global mean. Then, the classification fitness index equals the total sum of squared deviations minus the difference between the total sum of squared deviations within each category, divided by the total sum of squared deviations. The classification fitness index ranges from 0 to 1; a higher value indicates greater homogeneity within the partitions and greater differences between partitions, resulting in better classification performance.

[0127] Iterate through all possible combinations of boundaries, calculate the classification suitability index for each combination, and finally select the boundary that maximizes the index as the final target classification boundary.

[0128] After determining the new target classification boundaries, each cell in the updated preliminary aquifer water-bearing spatial distribution map is traversed, its water-bearing value is read, and the corresponding water-bearing level label is assigned according to the interval in which the value is located.

[0129] For example, pixels with water abundance values ​​below the first threshold can be identified as weakly water-rich areas and assigned the grade "A"; those between the two thresholds can be identified as moderately water-rich areas and assigned the grade "B"; and those above the second threshold can be identified as strongly water-rich areas and assigned the grade "C". In this way, each pixel gains a discrete grade attribute in addition to its original numerical attribute.

[0130] Then, spatial adjacency analysis can be performed on all pixels with the same level label, and spatially adjacent pixels with the same label can be grouped into the same water-rich level block.

[0131] Finally, the boundary geometry information of each newly generated water-bearing level block is extracted to obtain the updated water-bearing zoning results of the aquifer.

[0132] Among them, the updated aquifer water-bearing zoning results have improved in accuracy and reliability compared to the initial zoning results because they incorporate newly revealed measured data during the mining process and are reflected through model optimization. This can provide a more accurate basis for water prevention and control for subsequent mining activities.

[0133] The method provided in this application receives a multi-source heterogeneous raw dataset of the aquifer in the roof of the target coal seam and performs data reconstruction processing to generate a reconstructed geological data volume with spatial correlation. It calls a pre-trained aquifer water-bearing identification model to perform spatial feature learning processing on the reconstructed geological data volume to obtain a preliminary spatial distribution map of aquifer water-bearing. Then, based on the trend of water-bearing numerical distribution, it performs spatial division to generate aquifer water-bearing zoning results. When acquiring supplementary geological measurement data revealed during mining, it uses the data to iteratively optimize the aquifer water-bearing identification model and update the zoning results. The fusion of multi-source heterogeneous data improves the comprehensiveness of the data, the spatial feature learning ability of the artificial intelligence model portrays the high-dimensional coupling relationship between multiple factors under complex geological conditions, and the iterative optimization mechanism realizes dynamic updates based on real-time geological data, thereby improving the prediction accuracy, prediction robustness, and real-time response capability of aquifer water-bearing.

[0134] Optionally, after spatially dividing the target coal seam roof aquifer based on the numerical distribution trend of water-bearing properties in the preliminary aquifer water-bearing property spatial distribution map to generate the water-bearing property zoning result corresponding to the target coal seam roof aquifer, the method may further include:

[0135] S210. Extract the target water-bearing level blocks with the same water-bearing level identifier and the block boundary geometric information corresponding to the target water-bearing level blocks from the aquifer water-bearing zoning results.

[0136] After generating preliminary aquifer water-bearing zoning results, specific blocks can be selected from these zoning results for in-depth study according to the needs of subsequent analysis or engineering applications.

[0137] For example, to prevent the risk of sudden flooding, it may be necessary to focus on those blocks classified as "strongly water-rich". In practice, all blocks marked "strongly water-rich" can be selected from the "water-rich zoning" layer stored in the geodatabase using a query, and these blocks can be used as the target water-rich blocks.

[0138] For each selected target block, its associated boundary geometry information needs to be extracted from the database. This boundary geometry information is typically stored as vector polygons, defining the block's outline and extent in geographic space. The extracted boundary geometry information will serve as the basis for locating the block's spatial position and extracting its internal data in subsequent steps. This extraction process can be achieved using the spatial query function of Geographic Information System (GIS) software. By specifying attribute conditions such as a level identifier equal to a specific value, the software will return geometric objects that meet the criteria.

[0139] S220. Based on the block boundary geometry information, delineate the spatial range corresponding to the target water-bearing level block in the preliminary aquifer water-bearing spatial distribution map, and extract the water-bearing values ​​recorded at all spatial locations within the spatial range from the preliminary aquifer water-bearing spatial distribution map to generate a set of water-bearing values ​​within the block.

[0140] After obtaining the boundary geometry information of the target block, the spatial analysis function of the geographic information system software can be used to overlay this vector polygon onto the initial preliminary spatial distribution map of aquifer water-bearing capacity. Through spatial clipping, all pixels located inside the boundary of the polygon on the distribution map can be precisely clipped out, thereby delineating an area that completely corresponds to the spatial extent of the block.

[0141] Spatial clipping is a common spatial analysis operation that cuts a raster dataset using a vector polygon feature as a boundary, retaining only the raster portion within the boundary. Then, it's necessary to iterate through every cell within this clipped region. Since the clipped result is still a raster dataset, all cells can be traversed by reading an array of their cell values.

[0142] For each cell, its attribute value is read, which is the water-bearing value predicted by the model and recorded for that cell. These water-bearing values ​​read from all cells within the block are aggregated to form a set representing the distribution of water-bearing values ​​within the block, i.e., the block-internal water-bearing value set. The number of values ​​in the block-internal water-bearing value set is equal to the number of raster cells contained in the block.

[0143] S230. Perform central trend analysis on the set of water-bearing values ​​within the block to determine the representative water-bearing values ​​of the target water-bearing level block.

[0144] A water abundance level block contains numerous pixels, each with a water abundance value. While these values ​​belong to the same level range, they still exhibit some dispersion. To quantify and represent the water abundance level of the entire block, a central tendency analysis (CTA) is needed on the set of water abundance values ​​within the block. CTA aims to find a statistic that reflects the location or typical level of the data center. Since the median method may be used later, the data needs to be sorted first.

[0145] S231. Arrange all water-bearing values ​​in the set of water-bearing values ​​within the block in ascending order to generate an ordered sequence of water-bearing values.

[0146] First, all values ​​in the set of water-rich values ​​within the block can be sorted. For example, a quicksort algorithm can be used to arrange these disordered values ​​in ascending order of their numerical value, forming an ordered sequence.

[0147] After sorting, the original set of values ​​becomes a sequence arranged from minimum to maximum. For example, if a block contains multiple cells, their water abundance values, after sorting, will result in a sequence where the first element is the minimum, the last element is the maximum, and the elements in between increase sequentially.

[0148] S232. Identify the target water-bearing value located in the middle of the ordered water-bearing value sequence, and determine the identified target water-bearing value as the block representative water-bearing value of the target water-bearing level block.

[0149] After obtaining an ordered sequence, the value located in the middle of the sequence can be identified, such as the median. This median can effectively represent the general level of the entire dataset. Therefore, this identified median value can be determined as the representative water-bearing value for the target water-bearing level block. The representative water-bearing value is a typical value of the water-bearing capacity within that block.

[0150] S240. Generate block attribute information of the target water-rich level block based on the water-rich value represented by the block and the water-rich level identifier corresponding to the target water-rich level block.

[0151] After calculating the representative water-rich value of the block, this quantitative value can be combined with the block's original qualitative grade identifier to form a more complete attribute description of the block.

[0152] The block attribute information can contain multiple fields. For example, it can include a "water abundance level" field with a value of "strong water abundance"; and a new "representative water abundance value" field with a calculated median value.

[0153] In this way, the block is not only a surface with a grade label, but also a comprehensive information body that has both qualitative grades and quantitative indicators.

[0154] S250. The block attribute information and the block boundary geometric information are associated and stored to generate an aquifer water-bearing zoning result containing attribute annotations.

[0155] In this step, the newly generated block attribute information containing water-bearing values ​​can be associated with the previously stored boundary geometry information of the block and stored back in the geographic information database.

[0156] In other words, the existing water-rich partition layer can be updated by adding a new attribute field to the layer's data table and filling the calculated water-rich value into the record of the corresponding block.

[0157] In Geographic Information Systems (GIS), geometric and attribute information are linked through unique identifiers. By updating the attribute table, each block's geometry corresponds to a complete set of attributes. After this processing, the final aquifer water-bearing zoning delineation results contain richer information: geometric information describing the location and shape of the blocks, qualitative information describing the block's water-bearing level, and quantitative information describing the typical size of the block's water-bearing capacity.

[0158] The aforementioned partitioning results, which include attribute annotations, provide more accurate parameter support for subsequent work such as water inflow prediction, water control engineering design, and mine drainage capacity verification.

[0159] The method provided in this application embodiment can quantify the representative water-bearing value of each block by performing central tendency analysis on the internal values ​​of the divided water-bearing level blocks. This transforms the qualitative zoning results into results with quantitative attributes, improves the information content and practical value of the zoning results, and enables the zoning results to be directly used for subsequent hydrogeological calculations and engineering decisions.

[0160] Optionally, after step S250, the method may further include:

[0161] S310. During the mining of the aquifer on the roof of the target coal seam, the measured data of the supplementary water inflow points exposed within the spatial range corresponding to the target water-bearing level block are obtained. The measured data of the supplementary water inflow points includes the spatial coordinates of the water inflow points and the measured water-bearing data of the water inflow points recorded at the spatial coordinates of the water inflow points.

[0162] As mining operations continue, when tunnel excavation or face mining advances into a pre-defined target water-rich area, such as a "highly water-rich" block, new groundwater outbursts may be revealed. Once an outburst occurs, on-site engineers can immediately measure and record the data, obtaining the latest measured information. This latest data directly reflects the actual underground conditions and serves as a direct verification of previous predictions.

[0163] Supplementary measured data for the water inflow point may include, for example, the precise spatial coordinates of the water inflow point and measured water-bearing data obtained at that point through methods such as flow rate measurement. The precise spatial coordinates of the water inflow point can be obtained through a downhole measurement system; these coordinates need to be accurate to the meter level or even higher to ensure accurate location on the map. The measured water-bearing data can be obtained by measuring the stable water inflow at the water inflow point using a triangular weir or flow meter, while simultaneously measuring the head pressure at that point, and calculating the data in conjunction with the structural parameters of the discharge orifice.

[0164] Supplementing the measured data of the water inflow points has high reliability and can be used for model validation and correction to further improve the model's predictive accuracy.

[0165] S320. Based on the spatial coordinates of the water inflow point, determine the spatial water-bearing value of the spatial location point corresponding to the spatial coordinates of the water inflow point from the set of water-bearing value values ​​within the block.

[0166] After obtaining the precise spatial coordinates of the water inrush point, these coordinates can be used to locate it on the previously generated preliminary spatial distribution map of the aquifer's water-bearing capacity.

[0167] The preliminary spatial distribution map of aquifer water-bearing capacity is a georeferenced raster image, with each cell having a defined geographical extent. By combining the coordinates of the water inflow point with the geographic transformation information of the raster image, it is possible to calculate which specific cell the coordinate point falls on.

[0168] For example, given the coordinates of the top-left corner of an image and the pixel size, the row and column numbers of the pixel can be calculated using the coordinate difference. Once the row and column numbers are determined, the water-bearing value initially predicted by the aquifer water-bearing identification model can be retrieved from the pixel's attributes. This water-bearing value represents the model's prediction of the water-bearing capacity of that specific location before mining exposure.

[0169] S330. Compare and analyze the water-bearing value of the spatial point with the measured water-bearing data of the water-bearing point to determine the degree of numerical deviation between the water-bearing value of the spatial point and the measured water-bearing data of the water-bearing point.

[0170] The model predictions read from the distribution map in step S320 are compared and analyzed with the water-bearing data of the inflow points measured in the field in step S310. The purpose of the comparison is to evaluate the accuracy of the model predictions. The degree of numerical deviation between the two can be calculated.

[0171] One possible approach is to calculate the absolute deviation, which involves subtracting the measured value from the predicted value and then taking the absolute value. Another possible approach is to calculate the relative deviation, which involves dividing the absolute deviation by the measured value and then multiplying by 100%, expressing it as a percentage.

[0172] The numerical deviation quantitatively characterizes the difference between the model's prediction at a specific point and the objective reality. The smaller the deviation value, the more accurate the model's prediction; the larger the deviation value, the more likely the model's prediction at that point is flawed.

[0173] S340. Based on the comparison result between the numerical deviation degree and the preset deviation tolerance range, generate a block reliability verification conclusion for the target water-rich level block.

[0174] In order to determine whether the deviation calculated in step S330 is within an acceptable range, a deviation tolerance range can be preset. The deviation tolerance range can be preset according to factors such as engineering requirements and safety levels, and this application does not impose any restrictions on it. For example, it can be set that the absolute deviation must be less than a certain threshold, or the relative deviation must be less than a certain percentage.

[0175] Next, the calculated numerical deviation can be compared with a preset tolerance range. If the deviation is within the tolerance range, for example, if the relative deviation is less than a set percentage, a positive block reliability verification conclusion can be generated, indicating that the model's prediction results within the block match the actual situation well, and the prediction for the block is reliable.

[0176] Conversely, if the deviation exceeds the tolerance range, such as a relative deviation greater than a set percentage, a negative block reliability verification conclusion can be generated, indicating that the prediction results of the model within the block differ from the actual situation, and the prediction results of the block need to be corrected.

[0177] S350. When the reliability verification conclusion of the block indicates that the numerical deviation exceeds the preset deviation tolerance range, the representative water-bearing value of the block is corrected based on the measured water-bearing data of the water inrush point to generate a corrected representative water-bearing value of the block.

[0178] When the verification conclusion of step S340 indicates that the deviation exceeds the tolerance range, that is, the block prediction result is unreliable, it is necessary to use the newly acquired measured data to correct the original block representative water-rich value.

[0179] Specifically, new measured data can be incorporated into the existing dataset as additional information, and representative values ​​can be recalculated. For example, measured water-bearing data from water inflow points can be added as new values ​​to the intra-block water-bearing value set generated in step S220, thus forming an updated intra-block water-bearing value set containing measured information. Then, the central tendency analysis process in step S230 is re-executed on this updated set, that is, all values ​​are sorted again, and a new median is found.

[0180] Because new measured values ​​have been incorporated, the sorting sequence of the entire dataset will change, and the median position may shift accordingly. Therefore, the new median may differ from the original median. This new median, because it incorporates real measured data, is more representative of the true water-bearing level of the block than the original median, and is therefore determined as the corrected representative water-bearing value for the block.

[0181] S360. Update the corrected block representative water-bearing value to the block attribute information to generate an aquifer water-bearing zoning result containing the corrected attribute annotation.

[0182] Finally, the corrected water-bearing value of the block calculated in step S350 is updated in the attribute information of the target water-bearing level block in the geographic information database. This may include modifying the "representative water-bearing value" field of the corresponding block record in the "water-bearing zoning" layer, replacing the old value with the new corrected value.

[0183] The update operation can be completed using database update statements, specifying the layers and records to be updated and assigning the new values ​​to the corresponding fields. After such an update, the attribute information of the block becomes more accurate, including the corrected attribute labels. At the same time, this correction also provides more reliable basic data for subsequent model iterations and optimizations. The entire partitioning result has thus received localized and refined correction, making it closer to the actual geological conditions.

[0184] The method provided in this application verifies and corrects the divided blocks by using measured data of the exposed water inflow points during the mining process. This allows for dynamic updates to the quantitative attributes of the zoning results, enabling the zoning results to continuously self-calibrate and improve as more mining information is revealed. This provides a more reliable basis for subsequent water control decisions in the mining area.

[0185] Figure 2 This is a schematic diagram of the structure of an aquifer water-bearing zoning system 100 based on artificial intelligence, provided as an embodiment of this application. Figure 2As shown, the processor 120 can be used in an AI-based aquifer water-rich zoning system 100 and to perform the functions in this invention.

[0186] The AI-based aquifer water-bearing capacity zoning system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the AI-based aquifer water-bearing capacity zoning method of this invention. Although only one server is shown in this invention, for convenience, the functions described in this invention can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0187] For example, the AI-based aquifer water-bearing zoning system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the AI-based aquifer water-bearing zoning system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention can be implemented according to these program instructions. The AI-based aquifer water-bearing zoning system 100 also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0188] For ease of explanation, only one processor is described in the AI-based aquifer water-bearing zoning system 100. However, it should be noted that the AI-based aquifer water-bearing zoning system 100 of this invention may also include multiple processors. Therefore, the steps executed by one processor described in this invention may also be executed jointly by multiple processors or individually. For example, if the processor of the AI-based aquifer water-bearing zoning system 100 executes steps A and B, it should be understood that steps A and B may also be executed jointly by two different processors or individually by one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor jointly execute steps A and B.

[0189] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.

[0190] Finally, it should be noted that the above-disclosed embodiments are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for zoning aquifer water-bearing capacity based on artificial intelligence, characterized in that, The method includes: Receive a multi-source heterogeneous raw dataset corresponding to the aquifer on the roof of the target coal seam, wherein the multi-source heterogeneous raw dataset includes structural geological raw data, stratigraphic lithology raw data, and geophysical raw data; The multi-source heterogeneous original dataset is reconstructed to generate a reconstructed geological data volume corresponding to the target coal seam roof aquifer. The reconstructed geological data volume includes a spatial distribution layer of structural elements, a spatial distribution layer of lithological elements, and a spatial distribution layer of geophysical attributes with spatial correlation. The aquifer water-bearing property identification model, which has been pre-trained, is invoked to perform spatial feature learning processing on the reconstructed geological data volume. The aquifer water-bearing property identification model performs feature mapping operations at each spatial location of the reconstructed geological data volume to obtain a preliminary spatial distribution map of the aquifer water-bearing property corresponding to the target coal seam roof aquifer. Based on the distribution trend of water-bearing value in the preliminary aquifer water-bearing spatial distribution map, the target coal seam roof aquifer is spatially divided to generate the water-bearing zoning result of the target coal seam roof aquifer. When supplementary geological measurement data of the aquifer exposed during the mining process are obtained from the target coal seam roof aquifer, the aquifer water-bearing identification model is iteratively optimized using the supplementary geological measurement data, and the aquifer water-bearing zoning result is updated based on the optimized aquifer water-bearing identification model.

2. The aquifer water-bearing zoning method based on artificial intelligence according to claim 1, characterized in that, The process involves calling a pre-trained aquifer water-bearing capacity identification model to perform spatial feature learning on the reconstructed geological data volume. The model then performs feature mapping operations at various spatial locations within the reconstructed geological data volume to obtain a preliminary spatial distribution map of the water-bearing capacity of the target coal seam roof aquifer, including: The reconstructed geological data volume is segmented according to spatial coordinate points to obtain multiple spatial point feature vectors. Each spatial point feature vector contains the values ​​of structural elements, lithological elements, and geophysical attributes at the corresponding coordinate point. The spatial point feature vector is input into the aquifer water-bearing property identification model to determine the final leaf node to which the spatial point feature vector belongs in the decision tree group structure of the aquifer water-bearing property identification model, and the water-bearing property reference value stored in the final leaf node, so as to obtain a preliminary discrimination result. The preliminary discrimination results output by each decision tree in the aquifer water-bearing identification model are aggregated, and the statistical average of the preliminary discrimination results output by all decision trees is taken as the water-bearing discrimination output value corresponding to the spatial point feature vector. The water-bearing discrimination output value is backfilled according to the spatial coordinate points corresponding to the water-bearing discrimination output value to generate a two-dimensional water-bearing numerical distribution matrix that covers the reconstructed geological data volume space. The two-dimensional water-bearing numerical distribution matrix is ​​converted into a visualized raster image to obtain the preliminary spatial distribution map of water-bearing properties of the aquifer.

3. The aquifer water-bearing zoning method based on artificial intelligence according to claim 2, characterized in that, The step involves inputting the spatial point feature vector into the aquifer water-bearing capacity identification model, determining the final leaf node to which the spatial point feature vector belongs in the decision tree group structure of the aquifer water-bearing capacity identification model, and the water-bearing capacity reference value stored in the final leaf node, to obtain a preliminary discrimination result, including: The spatial point feature vectors are respectively input into each decision tree in the decision tree group structure, and a layer-by-layer matching process is performed on each decision tree starting from the root node. Based on the split feature dimension identifier stored in the current decision tree node, the corresponding feature value is extracted from the feature vector of the spatial point, and the feature value is compared with the split threshold stored in the current decision tree node to determine the next branch child node; After updating the next branch sub-node to the new current decision tree node, the layer-by-layer matching process is repeated until the leaf node of the decision tree is reached, and the leaf node is determined as the final leaf node to which the spatial point feature vector belongs on the decision tree. The average value of the water abundance of the samples in the training sample set covered by the final assigned leaf node is read from the data stored in the final assigned leaf node, and the average value is used as the preliminary discrimination result of the current decision tree outputting the feature vector of the spatial point.

4. The aquifer water-bearing zoning method based on artificial intelligence according to claim 1, characterized in that, The process of spatially dividing the aquifer roof of the target coal seam based on the numerical distribution trend of water-bearing properties in the preliminary aquifer water-bearing property spatial distribution map, and generating the aquifer water-bearing property zoning result corresponding to the aquifer roof of the target coal seam, includes: Extract the water-bearing values ​​of all spatial locations from the preliminary aquifer water-bearing spatial distribution map, and arrange the extracted water-bearing values ​​in ascending order to generate a water-bearing value sequence. The degree of difference between adjacent water-bearing values ​​in the water-bearing value sequence is calculated, and multiple water-bearing values ​​are selected from the water-bearing value sequence based on the change characteristics of the degree of difference to generate a candidate set of hierarchical boundaries. Based on the numerical intervals corresponding to the candidate values ​​of the classification boundary included in the classification boundary candidate set, the water-bearing numerical discrete characteristics within each numerical interval, and the water-bearing numerical difference characteristics between each numerical interval, the target classification boundary is determined from the classification boundary candidate set. Based on the target classification boundary, the water-bearing level of each spatial location point in the preliminary aquifer water-bearing spatial distribution map is determined, and the spatial location points with water-bearing values ​​in different value ranges are assigned corresponding water-bearing level labels. Spatial adjacency analysis is performed on spatial locations with the same water abundance level identifier, and spatial locations that are adjacent in location and have the same water abundance level identifier are grouped into the same water abundance level block. Generate block boundary geometric information for each of the water-bearing grade blocks, and associate the block boundary geometric information with the corresponding water-bearing grade identifier and store it as the water-bearing zoning result of the aquifer.

5. The aquifer water-bearing zoning method based on artificial intelligence according to claim 1, characterized in that, When supplementary geological measurement data of the aquifer exposed during the mining process of the target coal seam roof is obtained, the aquifer water-bearing capacity identification model is iteratively optimized using the supplementary geological measurement data, and the aquifer water-bearing capacity zoning result is updated based on the optimized aquifer water-bearing capacity identification model, including: The newly revealed spatial location points, newly revealed structural geological data, newly revealed stratigraphic lithology data, newly revealed geophysical data, and newly revealed measured water-bearing data are extracted from the supplementary geological measurement data. Perform data reconstruction processing on the newly revealed structural geological data, the newly revealed stratigraphic lithology data, and the newly revealed geophysical data to generate a new revealed reconstructed geological data body with the same data structure as the reconstructed geological data body; The newly revealed reconstructed geological data is input into the aquifer water-bearing identification model for water-bearing prediction processing to obtain the predicted water-bearing value corresponding to the newly revealed spatial location point. The model deviation of the aquifer water-bearing identification model is determined based on the deviation between the predicted water-bearing value and the newly revealed measured water-bearing data. The decision tree group structure inside the aquifer water-bearing identification model is corrected according to the model deviation. The water-bearing reference data stored in the leaf nodes of the decision tree group structure is updated, and the feature selection criteria used for path division in the decision tree nodes are adjusted so that the difference between the predicted water-bearing value output by the corrected aquifer water-bearing identification model at the newly revealed spatial location point and the newly revealed measured water-bearing data is less than a preset difference threshold, thus obtaining the optimized aquifer water-bearing identification model. The reconstructed geological data volume is re-inputted into the optimized aquifer water-bearing identification model for water-bearing prediction processing to obtain an updated preliminary aquifer water-bearing spatial distribution map. Based on the updated preliminary aquifer water-bearing spatial distribution map, spatial division processing is performed to obtain the updated aquifer water-bearing zoning results.

6. The aquifer water-bearing capacity zoning method based on artificial intelligence according to claim 5, characterized in that, The process involves correcting the decision tree group structure within the aquifer water-bearing capacity identification model based on the model deviation, updating the water-bearing capacity reference data stored in the leaf nodes of the decision tree group structure, and adjusting the feature selection criteria used for path partitioning in the decision tree nodes. This ensures that the difference between the predicted water-bearing capacity value output by the corrected aquifer water-bearing capacity identification model at the newly revealed spatial location point and the newly revealed measured water-bearing capacity data is less than a preset difference threshold, resulting in an optimized aquifer water-bearing capacity identification model. This includes: The leaf node that the newly revealed reconstructed geological data volume finally reaches is determined on each decision tree within the aquifer water-bearing identification model; Obtain the water-bearing reference data currently stored in the final reached leaf node, and determine the preliminary prediction results of each decision tree for the output of the newly revealed reconstructed geological data volume based on the obtained water-bearing reference data stored in the leaf nodes of each decision tree. The preliminary prediction results output by all the decision trees are comprehensively calculated to obtain the comprehensive predicted water-bearing value corresponding to the newly revealed reconstructed geological data volume; The model bias is determined based on the difference between the comprehensive predicted water-bearing value and the newly revealed measured water-bearing data; Based on the model bias, the degree of similarity between the preliminary prediction results output by each decision tree and the newly revealed measured water-bearing data, the correction component of each decision tree is determined. The water-bearing reference data stored in the leaf node of the newly revealed reconstructed geological data volume finally reached on each decision tree is adjusted according to the correction component. After adjusting all the decision trees that need adjustment, the optimized aquifer water-bearing identification model is obtained based on the overall structure of the adjusted decision tree group.

7. The aquifer water-bearing zoning method based on artificial intelligence according to claim 1, characterized in that, After spatially dividing the target coal seam roof aquifer based on the water-bearing value distribution trend in the preliminary aquifer water-bearing spatial distribution map, and generating the aquifer water-bearing zoning result corresponding to the target coal seam roof aquifer, the process further includes: Extract the target water-bearing level blocks with the same water-bearing level identifier and the block boundary geometric information corresponding to the target water-bearing level blocks from the aquifer water-bearing zoning results. Based on the geometric information of the block boundary, the spatial range corresponding to the target water-bearing level block is delineated in the preliminary aquifer water-bearing spatial distribution map, and the water-bearing values ​​recorded by all spatial location points within the spatial range are extracted from the preliminary aquifer water-bearing spatial distribution map to generate a set of water-bearing values ​​within the block. A central trend analysis is performed on the set of water-bearing values ​​within the block to determine the representative water-bearing values ​​of the target water-bearing level block. Based on the water-bearing value represented by the block and the water-bearing level identifier corresponding to the target water-bearing level block, the block attribute information of the target water-bearing level block is generated. The block attribute information is associated and stored with the block boundary geometry information to generate aquifer water-bearing zoning results containing attribute annotations.

8. The aquifer water-bearing zoning method based on artificial intelligence according to claim 7, characterized in that, The process of performing central trend analysis on the set of water-bearing values ​​within the block to determine the representative water-bearing values ​​of the target water-bearing level block includes: Arrange all water-bearing values ​​in the set of water-bearing values ​​within the block in ascending order to generate an ordered sequence of water-bearing values. Identify the target water-bearing value located in the middle of the ordered water-bearing value sequence, and determine the identified target water-bearing value as the block representative water-bearing value of the target water-bearing level block.

9. The aquifer water-bearing capacity zoning method based on artificial intelligence according to claim 7, characterized in that, After associating and storing the block attribute information with the block boundary geometric information to generate an aquifer water-bearing zoning result containing attribute annotations, the method further includes: During the mining of the aquifer on the roof of the target coal seam, measured data of supplementary water inflow points exposed within the spatial range corresponding to the target water-bearing grade block are obtained. The measured data of supplementary water inflow points includes the spatial coordinates of the water inflow points and the measured water-bearing data of the water inflow points recorded at the spatial coordinates of the water inflow points. Based on the spatial coordinates of the water inflow point, determine the spatial water-bearing value of the spatial location point corresponding to the spatial coordinates of the water inflow point from the set of water-bearing value values ​​within the block; The water-bearing capacity values ​​at the spatial points are compared and analyzed with the measured water-bearing capacity data at the water inflow points to determine the degree of numerical deviation between the water-bearing capacity values ​​at the spatial points and the measured water-bearing capacity data at the water inflow points. Based on the comparison between the numerical deviation and the preset deviation tolerance range, a block reliability verification conclusion is generated for the target water-rich level block. When the reliability verification conclusion of the block indicates that the numerical deviation exceeds the preset deviation tolerance range, the representative water-bearing value of the block is corrected based on the measured water-bearing data of the water inrush point to generate a corrected representative water-bearing value of the block. The corrected water-bearing value of the block is updated in the block attribute information to generate aquifer water-bearing zoning results containing the corrected attribute labels.

10. An aquifer water-bearing zoning system based on artificial intelligence, characterized in that, The device includes a processor and a computer-readable storage medium storing machine-executable instructions, which, when executed by a computer, implement the artificial intelligence-based aquifer water-rich zoning method according to any one of claims 1-9.

Images