A three-dimensional modeling method of a coal mining face with cross-sectional boundary resistance traction

Abstract

The application provides a three-dimensional modeling method of a coal mining face profile boundary resistance traction, and relates to the field of three-dimensional modeling of a coal mining face. The method comprises the following steps: collecting two-roadway geological profile data and mining face geological profile data of the coal mining face; pre-processing and feature extraction are performed on the two-roadway geological profile data and the mining face geological profile data to obtain a boundary feature point set; a cross-profile resistance constraint network is constructed according to the boundary feature point set; under the action of the resistance constraint network, a geological body surface point set is generated through an adaptive interpolation algorithm; and three-dimensional morphological fitting is performed on the geological body surface point set to obtain a three-dimensional model of the coal mining face. The application solves the problems of low complexity geological body form restoration degree and missing model attribute information, and the constructed three-dimensional model is not only complete and non-segmented, but also has accurate geometric form and comprehensive attribute information.

Description

Technical Field

[0001] This application relates to the field of three-dimensional modeling of coal mining faces, and in particular to a three-dimensional modeling method for coal mining faces based on profile boundary resistance traction. Background Technology

[0002] A coal mining face is a coal-bearing area within a mining field, designated according to the mining plan. Accessible roadways are excavated on both sides of its direction. The coal mining machine begins excavating from the end of the roadway and then continuously retreats until all mining is completed. To understand the three-dimensional distribution of the coal seam and geological structures ahead, coal production departments typically collect geological profiles of the two roadways beforehand and periodically collect geological profiles of the mining face. However, these geological profiles only provide a geometrically fragmented view of the coal seam and geological structure distribution, not a complete three-dimensional model, making it difficult to provide decision support for intelligent coal mining and mining volume prediction. Three-dimensional modeling technology reconstructs the aforementioned geological profiles into a complete and unfragmented three-dimensional model, providing decision support services for intelligent coal mining and mining volume prediction.

[0003] Chinese invention patent specification CN202311385994 discloses a modeling method for multiple coal seams and multiple working faces. This method involves drawing dip profiles in AutoCAD software, adjusting the order of the dip profiles in Rhino software, and finally using Rhino's loft command to obtain a 3D model of the coal seam working face. The core technology of this method originates from Rhino's lofting technology, but the specific method of lofting is not publicly disclosed. Furthermore, Rhino's lofting method is difficult to handle profile lofting involving geological bodies such as faults and folds. Chinese invention patent specification CN202311310249 discloses a modeling method for longwall mining faces. This method first extracts surface contour lines, generates surface surfaces in Rhino software, then intersects these surfaces with predefined cuboids to obtain a 3D geological model. Finally, it uses Grasshopper battery packs to parametrically generate coal seams and excavate roadways. This method only uses surface contour lines as data, and the resulting model is a simulation model, unable to provide decision support services for intelligent coal mining and mining volume prediction.

[0004] In summary, existing profile-based modeling techniques are insufficient for handling geological bodies containing faults and folds, and cannot provide decision support services for intelligent coal mining and mining volume prediction. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a three-dimensional modeling method for coal mining faces based on profile boundary resistance traction, which at least partially solves the problems existing in the prior art.

[0006] In a first aspect of this application, a three-dimensional modeling method for coal mining faces induced by profile boundary resistance is provided, the method comprising the following steps:

[0007] Collect geological profile data of the two roadways and the mining face of the coal mining face;

[0008] Preprocessing and feature extraction were performed on the geological profile data of the two tunnels and the geological profile data of the mining face to obtain the set of boundary feature points;

[0009] Based on the set of boundary feature points, a cross-section resistance constraint network is constructed; wherein, the resistance constraint network includes cross-section resistance correlation lines and cross-section resistance transmission rules.

[0010] Under the influence of the resistance constraint network, an adaptive interpolation algorithm is used to generate a set of points on the surface of the geological body.

[0011] A three-dimensional morphological fitting was performed on the point set on the surface of the geological body to obtain a three-dimensional model of the coal mining face.

[0012] This application has at least the following beneficial effects:

[0013] The 3D modeling method for coal mining faces based on profile boundary resistance provided in this application firstly acquires comprehensive multi-source, multi-format data, including borehole data, seismic exploration data, and measured roadway data, by collecting geological profile data from the two roadways and the mining face of the coal mining face. This avoids the loss of model information caused by a single data source. Next, through refined preprocessing and feature extraction, not only are outliers automatically removed and minor anomalies corrected using clustering algorithms, but feature extraction models also accurately identify difficulties such as blurred roof and floor boundaries, discontinuous fault traces, and concealed fold axes. Complete boundary feature parameters and key nodes are extracted simultaneously. Then, a cross-profile resistance constraint network is constructed based on this feature point set. Quantitative resistance coefficient assignment rules, clear constraint direction definitions, and feature parameter similarity matching achieve accurate association of similar boundaries. Combined with distance-weighted and adjacent-cooperative resistance transmission rules, and by setting resistance abrupt or gradual change nodes for faults and folds, this effectively solves the problems of inaccurate cross-profile boundary matching and resistance transmission. The system addresses the lack of handling of regular and complex geological bodies, ensuring in-depth collaborative constraints of multi-profile data and avoiding boundary drift and geometric fragmentation during modeling. Subsequently, under the action of this resistance constraint network, an adaptive interpolation algorithm is used to design a three-level adaptive density grid based on profile distribution and key nodes. The grid nodes are associated with resistance constraint information, a resistance traction correction term is introduced, and a specific interpolation rule is designed for complex geological bodies. This achieves a balance between accuracy and efficiency, and ensures that the interpolation results closely match the real geological boundaries, avoiding morphological distortion caused by traditional uniform grids and unconstrained interpolation. Finally, through targeted 3D morphological fitting, specific fitting algorithms are applied to the coal seam roof and floor, faults, and fold axial surfaces, and thickness calibration, morphological integrity checks, and attribute mapping are performed to generate complete 3D models. This solves the problems of low morphological fidelity and missing model attribute information for complex geological bodies. The constructed 3D model is not only complete rather than fragmented, but also has accurate geometry and comprehensive attribute information, providing reliable decision support services for intelligent coal mining and mining volume prediction. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating a three-dimensional modeling method for coal mining faces based on profile boundary resistance provided in this application embodiment. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] It should be noted that the following description covers various aspects of embodiments within the scope of the appended claims. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this application, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0019] Please refer to Figure 1 As shown, an embodiment of this application provides a three-dimensional modeling method for coal mining faces with profile boundary resistance traction. The method includes the following steps:

[0020] S100 collects geological profile data of the two roadways and the mining face of the coal mining face.

[0021] Specifically, the geological profiles of the two roadways are strike profiles, and the geological profiles of the mining face are dip profiles. The geological profile data of the two roadways and the mining face include borehole data, seismic exploration data, and roadway measured data; the data formats cover CSV (coordinate data), BMP (profile image), and DXF (CAD format profile).

[0022] S200 involves preprocessing and feature extraction of geological profile data from the two tunnels and the mining face to obtain a set of boundary feature points.

[0023] Specifically, preprocessing includes preliminary noise filtering and outlier clustering and removal; among which, preliminary noise filtering includes:

[0024] S210, noise filtering is performed on the geological profile data of the two tunnels and the geological profile data of the mining face using Gaussian filtering.

[0025] Here, a Gaussian filtering algorithm is used to smooth the profile image data. For the coordinate data, the standard deviation of the neighborhood of each point is calculated, and discrete points with large standard deviations are removed.

[0026] Outlier clustering removal includes:

[0027] S220, the noise-filtered geological profile data of the two tunnels and the geological profile data of the mining face are converted into two-dimensional feature vectors, wherein the two-dimensional feature vectors have the distance along the profile length and the vertical depth as feature dimensions.

[0028] Here, vertical depth refers to the vertical depth coordinate of each data point in the geological profile relative to a preset reference surface, and is the core dimension reflecting the vertical positional relationship of the strata. Vertical depth is the Z-axis component in three-dimensional coordinates, and its value is determined by the following rules: with the preset reference surface (such as the mean sea level of the mining area or the mining horizontal reference surface) as the zero point, Z takes a positive value when the data point is below the reference surface, and the value is equal to the vertical distance from the point to the reference surface.

[0029] First, the initially filtered coordinate data is converted into a two-dimensional feature vector. (Because the profile data has continuity in the direction of orientation or dip, the distance along the profile length and the vertical depth are taken as the feature dimensions. That is, the feature vector of each data point is (distance along the profile length and vertical depth), which simplifies the calculation while retaining the core spatial features.)

[0030] S230. Based on the two-dimensional feature vector and the k-means clustering algorithm, the first cluster, the second cluster, and the third cluster are determined. The intra-cluster density, dispersion, and distance from other clusters of the first cluster, the second cluster, and the third cluster are all different.

[0031] Step S230 further includes:

[0032] S231, randomly select one two-dimensional feature vector from several two-dimensional feature vectors as the candidate first cluster center.

[0033] S232, calculate the neighborhood density of the candidate first cluster center.

[0034] S233, if the neighborhood density is greater than the preset normal neighborhood density threshold, the candidate center is determined as the first cluster center.

[0035] S234, calculate the Euclidean distance from each two-dimensional feature vector except the first cluster center to the first cluster center.

[0036] S235. Based on the Euclidean distance from each two-dimensional feature vector other than the first cluster center to the first cluster center, the roulette wheel selection method is used to determine the candidate second cluster center.

[0037] S236, if the distance from the candidate second cluster center to the first cluster center is greater than the first preset distance threshold, then the candidate second cluster center is determined as the second cluster center.

[0038] S237, calculate the Euclidean distance from each two-dimensional feature vector other than the first cluster center and the second cluster center to the second cluster center.

[0039] S238. Based on the Euclidean distance from each two-dimensional feature vector other than the first and second cluster centers to the second cluster center, the roulette wheel selection method is used to determine the candidate third cluster center.

[0040] S239, if the distance from the candidate third cluster center to the first cluster center is greater than the second preset distance threshold, and the distance to the second cluster center is greater than the third preset distance threshold, then the candidate third cluster center is determined as the third cluster center.

[0041] Here, one point is randomly selected from all data points as a candidate center; the neighborhood density of the candidate center is calculated. If the neighborhood density of the candidate center is greater than a preset normal neighborhood density threshold, then the candidate center is determined as the first cluster center (i.e., it is necessary to ensure that the first cluster center is a normal cluster center). Then, the Euclidean distance from all remaining points to the first cluster center is calculated, and all Euclidean distances are normalized to obtain the first distance weight for each point. The first distance weight is used as the probability that each point will be selected as the second cluster center. The second cluster center is selected using a roulette wheel selection method. A random number between 0 and 1 is generated, and the probabilities of each point are accumulated. When the accumulated sum is equal to or greater than the random number, the corresponding point is selected as a candidate second cluster center. Then, the rationality of the candidate second cluster center is verified: the distance from the candidate second cluster center to the first cluster center is calculated. If the distance from the candidate second cluster center to the first cluster center is greater than the first preset distance threshold, then the candidate second cluster center is determined as the second cluster center. At this time, the second cluster center can only be determined to be the cluster center of abnormal points, and cannot be determined to be the cluster center of mildly abnormal points or the cluster center of severely abnormal points.

[0042] Next, the distances from all remaining points to the first and second cluster centers are calculated, and the minimum distance between them is taken as the Euclidean distance for each point. The Euclidean distances are then normalized to obtain a second distance weight for each point. This second distance weight is used as the probability that each point will be selected as the third cluster center. A roulette wheel selection method is used to select the third cluster center: a random number between 0 and 1 is generated, and the probabilities of each point are accumulated. When the accumulated sum is equal to or greater than the random number, the corresponding point is selected as a candidate third cluster center. Then, the rationality of the candidate third cluster center is verified: the distances from the candidate third cluster center to the first and second cluster centers are calculated. If the distance from the candidate third cluster center to the first cluster center is greater than a second preset distance threshold, and the distance from the candidate third cluster center to the second cluster center is greater than a third preset distance threshold, then the candidate third cluster center is determined as the third cluster center. The Euclidean distances from each data point to the first, second, and third cluster centers are calculated, and the point is assigned to the cluster containing the smallest distance. Recalculate the center of each cluster (the mean of all two-dimensional feature vectors within the cluster) and update the center of each cluster; until the change in the position of the cluster center is equal to or less than the preset change in the cluster; the iteration stops; the final cluster classification is as follows: the cluster with the highest neighborhood density and the lowest dispersion (lowest standard deviation) is the normal cluster; the cluster with moderate dispersion and close to the normal cluster is the mildly anomalous cluster; the cluster with the highest dispersion and the furthest distance from the first two clusters is the severely anomalous cluster.

[0043] It should be noted that the dispersion of mild anomalies is relative to the concentrated distribution of normal point clusters, not absolute dispersion. Their anomalies stem from small-amplitude and systematic deviations (such as instrument measurement accuracy errors or slight deviations in borehole depth), rather than random extreme errors. Therefore, they possess two key clustering conditions: Key clustering condition one: consistent anomaly degree, meaning that the deviation range of all mild anomalies from normal points is similar and much smaller than that of severe anomalies; Key clustering condition two: continuous spatial distribution, meaning that mild anomalies are mostly distributed along local areas of the profile (e.g., borehole data within a 100m range all show slight depth deviations), rather than being randomly scattered throughout the entire profile. These two characteristics make the spatial distance between mild anomalies much smaller than their distance from normal point clusters and severe anomaly point clusters, satisfying the core logic of K-Means clustering where nearest neighbors are grouped together, thus enabling the formation of independent mild anomaly point clusters.

[0044] In this embodiment, the distance along the profile length and vertical depth are used as the classification criteria for clusters. When all mild anomalies are not clustered in one place, or when there are multiple scattered severe anomalies, all mild anomalies are classified into the corresponding mild anomaly cluster, and all severe anomalies are classified into the corresponding severe anomaly cluster. That is, a mild anomaly cluster may contain several mild anomaly sub-clusters; each sub-cluster is a mild anomaly cluster. A severe anomaly cluster may contain several scattered severe anomalies. In other words, in this embodiment, the cluster setting of K=3 (three cluster centers) is for class distinction rather than quantity limitation.

[0045] S2310 Calculate the Euclidean distance from each two-dimensional feature vector to the three cluster centers except for the first, second, and third cluster centers, and assign the data points to the cluster center with the smallest distance to obtain the first cluster, the second cluster, and the third cluster.

[0046] S240, the clusters with the smallest intra-cluster density, the largest dispersion, and the largest distance from other clusters in the first, second, and third clusters are identified as the clusters to be removed;

[0047] S250 removes all two-dimensional feature vectors in the cluster to be removed, corresponding to the geological profile data of the two tunnels and the geological profile data of the mining face.

[0048] It should be noted that feature extraction includes:

[0049] S201, acquire the profile grayscale image, attitude heat map and geological attribute mapping map corresponding to the preprocessed geological profile data of the two roadways and the geological profile data of the mining face; among them, the attitude heat map is an image that converts the dip angle and strike angle of the strata into thermal values; the geological attribute mapping map is an image that converts lithology, density and wave velocity into multi-channel features.

[0050] S202 uses the grayscale profile image, occurrence heat map and geological attribute mapping map as input, and performs feature extraction according to the feature extraction model to obtain the feature parameters of the coal seam roof and floor boundaries, fault traces and fold axis traces. The feature parameters include the three-dimensional coordinates of the boundary points, strike angle, dip angle, curvature and specific parameters. The specific parameters are at least one of the following: coal seam thickness, fault drop, and inter-flange angle.

[0051] The feature extraction model employs a composite loss function consisting of weighted cross-entropy and morphological loss. The morphological loss includes top and bottom plate boundary smoothness loss, fault trace continuity loss, and fold axis symmetry loss.

[0052] Here, the feature extraction model adopts multimodal input, where the profile grayscale image reflects visual grayscale features; the attitude heat map is an image that converts the dip angle and strike angle of the strata into thermal values; highlighting the changes in the attitude of the strata (which can be used to solve the problem of fold axis plane identification); the geological attribute mapping map is a feature map that converts attributes such as lithology, density, and wave velocity into different channels, and uses lithological differences to assist in the identification of the top and bottom plates (which can be used to solve the problem of ambiguity of the top and bottom plates).

[0053] In one embodiment, the feature extraction model is a multimodal fusion U-Net model. The feature extraction model uses a convolutional attention module to weightedly fuse the three modal features, focusing on strengthening feature channels related to boundary recognition (such as attitude change channels and lithological difference channels). The network structure of the feature extraction model is optimized as follows: Encoder part: Add edge enhancement convolutional kernels (3×3 kernels) to enhance boundary gray-scale abrupt change features, while retaining deep semantic features (such as lithological distribution); Decoder part: Introduce a fault trace completion module, and use a bidirectional long short-term memory network to perform sequence modeling on the fault candidate regions output by the decoder, completing the discontinuous fault traces (solving the problem of discontinuous fault traces); Output layer design: Use a multi-task output head to output the top and bottom plate boundary map, fault trace map, and fold axial surface trace map respectively. The loss function design employs a composite loss function of weighted cross-entropy and morphological loss. Weighted cross-entropy assigns higher weights to imbalanced regions (such as fold axial surfaces with low pixel proportions) (weight coefficient = 1 / pixel proportion). Morphological loss adds smoothness loss (based on adjacent pixel gradients) to the top and bottom plate boundaries, continuity loss (based on trace segment spacing) to fault traces, and symmetry loss (based on the difference in attitude between the two fold flanks) to fold axial surfaces. Feature extraction model training strategy: Dataset composition: 1000+ sets of profile data under different geological conditions (including faults, folds, and normal strata) are collected, divided into a training set (800 sets), a validation set (100 sets), and a test set (100 sets). Each set includes a profile image, a set of coordinate points, and manually labeled boundaries (top and bottom plate, faults, fold axial surfaces) and a geological attribute table (lithology, density, wave velocity). Next, data augmentation was performed: to address the boundary blurring problem, grayscale stretching and random occlusion were used for enhancement: for the blurred areas of the top and bottom plates, grayscale contrast was improved by histogram equalization; for the discontinuous fault areas, some trace segments were randomly occluded (simulating actual acquisition gaps) to train the model's completion ability; at the same time, conventional enhancement methods such as rotation (±10°), scaling (0.8-1.2 times), and noise superposition (Gaussian noise) were added.

[0054] The model was trained using the dataset described above. The model output includes top and bottom plate boundary line maps, fault trace maps, and fold axial surface trace maps.

[0055] The boundary lines (top and bottom plates, faults, and fold axial surfaces) identified in the aforementioned top and bottom plate boundary line maps, fault trace maps, and fold axial surface trace maps are uniformly sampled at preset intervals to generate an initial feature point set (each point contains three-dimensional coordinates, its boundary type, and feature parameters). Then, based on the profile acquisition accuracy, the neighborhood radius and minimum number of points are set, and the initial feature point set is clustered, with each cluster representing a continuous boundary segment. The central feature point (the average coordinates of all points within the cluster) and boundary feature points (the points farthest from the center within the cluster) of each cluster are retained, while redundant points within the cluster (such as duplicate sampling points in continuous, gentle sections) are removed. This simplifies the feature point set.

[0056] Key point identification is performed on the simplified point set: By calculating the curvature of each feature point, several key nodes are determined based on the curvature threshold corresponding to each key node. Key node types include fault breakpoints (trajectory start and end points, inflection points), fold inflection points (points where the wing transitions to the axial plane), and abrupt change points on the top and bottom plates (points with a thickness variation coefficient > 0.3). The curvature threshold is different for each type of key node. For the selected key nodes, additional sampling points are added within a certain range before and after them to form an encrypted feature point set, ensuring the preservation of details in key areas. Finally, the optimized feature point set is stored according to boundary type, with each point accompanied by coordinates, feature parameters, and a key node label (yes or no).

[0057] In this embodiment, redundant points are first clustered and simplified, and then key nodes are filtered and encrypted using a curvature threshold.

[0058] S300, based on the set of boundary feature points, construct a cross-section resistance constraint network; wherein, the resistance constraint network includes cross-section resistance correlation lines and cross-section resistance transmission rules.

[0059] Specifically, step S300 also includes:

[0060] S310, determine the tangent direction and normal direction of each feature point in the boundary feature point set; wherein, the tangent direction is calculated based on the coordinates of adjacent feature points, the normal direction is perpendicular to the tangent direction and points into the geological body, the adjustment range of the feature point along the tangent direction is less than the preset tangent adjustment threshold, and the displacement constraint of the normal direction is 0.

[0061] Here, feature point adjustment refers to the coordinate fine-tuning mechanism in the subsequent interpolation process. Through the dual rules of limited adjustment in the tangent direction and prohibition of movement in the normal direction, the smooth continuity of the geological boundary is ensured, while the core morphology of the boundary (such as thickness and range) is preserved without distortion.

[0062] The boundaries within a single profile are divided into three core types: coal seam roof and floor boundaries, fault traces, and fold axial surface traces. The hierarchical relationship of each type of boundary is also marked (e.g., which fault structure a fault trace belongs to). Each boundary type has a corresponding preset resistance coefficient. For each feature point, the tangent direction (based on the coordinates of adjacent feature points) and normal direction (perpendicular to the tangent direction, pointing into the geological body) of its boundary are determined. Feature points are only allowed to be fine-tuned along the tangent direction, and drifting along the normal direction is prohibited (the displacement constraint in the normal direction is 0), ensuring that the boundary morphology does not fundamentally change.

[0063] S320 uses the geological profiles of the two roadways of the coal mining face as the reference end and the geological profile of the mining face as the intermediate node. Several similar boundaries are determined by matching the similarity of the feature parameters of the reference end and the intermediate node.

[0064] Here, the two roadway profiles are used as reference ends (end A and end B), and their X-axis coordinates are recorded (end A X=0, end B X=L, where L is the length of the working face); the profiles collected periodically from the mining face are used as intermediate nodes (C1, C2, ..., C...). n The data is sorted according to the acquisition time (i.e., the working face advance distance), and the X-axis coordinate of each intermediate node is recorded; n is the number of intermediate nodes.

[0065] S330 connects the feature points of the reference end and the intermediate node corresponding to the same boundary to obtain the cross-sectional resistance correlation line.

[0066] Here, based on boundary feature parameters (strike angle, dip angle, fault displacement, inter-laminar angle), a similarity matching algorithm is used to pair similar boundaries across different profiles. For example, the similarity of feature parameters between a boundary at end A and a boundary at intermediate node C1 is calculated. If the similarity is greater than or equal to a preset similarity threshold, they are considered the same boundary (e.g., traces of the same fault, roof and floor of the same coal seam), and a matching relationship is established. For successfully matched boundaries of the same type, corresponding feature points across different profiles are connected (key nodes are connected first) to form a cross-profile resistance correlation line.

[0067] S340, based on distance-weighted rules and adjacent cooperation rules, transmits resistance coefficients to construct a cross-section resistance constraint network; wherein, the adjacent cooperation rule is triggered when the distance between adjacent intermediate nodes is less than a preset node distance threshold.

[0068] Here, resistance transfer refers to the transfer of the boundary constraint strength (resistance coefficient) of the reference profiles (the two tunnel profiles at ends A and B) to the intermediate profiles (the profiles collected during mining) according to certain rules. At the same time, the constraint strength of adjacent intermediate profiles affects each other, ultimately achieving continuous, unified, and geologically consistent boundary constraints for all profiles. This avoids setting the resistance coefficient of the intermediate profiles arbitrarily and ensures that its constraint logic is consistent with that of the two reference profiles and adjacent intermediate profiles. Consequently, during subsequent interpolation modeling, there will be no breakage or drift in the geological boundaries.

[0069] Distance-weighted transfer based on the two reference profiles (core base transfer): This step first assigns a distance weight to the intermediate profile C. i The base value of the drag coefficient is determined based on the distance from the intermediate section to end A and end B. The closer the distance, the greater the impact of the drag coefficient on the corresponding base end.

[0070] The distance-weighted rule meets the following conditions:

[0071] ;

[0072] in, Let be the basic resistance coefficient of the i-th intermediate node; The resistance coefficients at the starting points of the geological profiles of the two tunnels; is the resistance coefficient at the end of the geological profile of the two roadways; L is the total length of the coal mining face. is the distance from the i-th intermediate node to the starting point of the geological profile of the two lanes.

[0073] (LX) i ) / L represents the intermediate section C i The percentage of distance to point A; the closer to point A, the larger this percentage, k A For C i The higher the influence weight, the better; intermediate section C i The percentage of the distance to B is X i / L: The closer to the B-end, the larger this percentage. B For C i The higher the influence weight, the better.

[0074] As an example: Assume the total length of the working face is L = 1000m, and the drag coefficient k of a fault characteristic point at end A is... A =0.9 (fault elevation difference H≥5m), k corresponding to the same fault characteristic point at end B. B =0.85 (the fault has a drop of 3-5m at end B); X1=300m in the intermediate section C1 (300m from end A, 700m from end B): k C1=0.9×(1000-300) / 1000+0.85×300 / 1000=0.9×0.7+0.85×0.3=0.63+0.255=0.885; It can be seen that C1 is closer to end A, and the drag coefficient is closer to k. A (0.9) conforms to the geological law that the fault displacement gradually decreases from end A to end B, and the resistance coefficient changes synchronously.

[0075] If the distance between two adjacent intermediate profiles (such as C1 and C2) is less than the preset profile distance threshold (indicating dense profile collection and gradual geological condition changes), the influence of the resistance coefficient of adjacent profiles needs to be added to the basic value in the first step to avoid abrupt changes in the resistance coefficient of adjacent profiles.

[0076] Specifically, taking the intermediate section C i For example, its adjacent preceding section is C. i-1 (Collection order in C) i Previously, the X coordinate was less than X. i ); using the aforementioned C i Basic resistance coefficient k Ci Multiply by its own base value weight (for example: 0.8), and add C. i-1 The final drag coefficient kC i-1 Multiply by the influence weight of adjacent profiles (for example: 0.2) to get C. i Final drag coefficient: k Ci ’ =k Ci ×0.8+k Ci-1 ×0.2.

[0077] This embodiment allows the drag coefficients of adjacent sections to pull against each other, as an example: C i The base value is 0.885, C i-1 The final value of C is 0.89. i The final value = 0.885 × 0.8 + 0.89 × 0.2 = 0.708 + 0.178 = 0.886, which is consistent with C. i-1 The difference is only 0.004, achieving a smooth transition in the drag coefficient.

[0078] Applicable scenarios: This step is triggered only when the distance between adjacent intermediate profiles is less than a preset profile distance threshold (e.g., a profile is collected at regular intervals during mining). If the distance is greater than or equal to the preset profile distance threshold (sparse profiles may indicate abrupt changes in geological conditions), this step is not performed to avoid incorrect transmission of the resistance coefficient in abruptly changed areas.

[0079] In summary, the core principles of the transfer process are: first, use the two reference profiles, namely A and B, as anchor points to ensure that the resistance coefficient of the intermediate profile does not deviate from the overall geological constraint framework; the farther away from a certain reference point, the weaker the influence of the resistance coefficient at that end, which conforms to the gradual change law of geological structure (for example, the fault drop and the tightness of folds usually do not change suddenly); and the densely collected intermediate profiles influence each other, strengthen the continuity of constraints, and avoid the fragmentation of the boundary morphology of adjacent profiles during modeling.

[0080] In addition, at the cross-section matching point of the fault trace, a resistance abrupt change marker is set, and the resistance correlation line on both sides of the abrupt change node is broken (resistance is not transmitted), to ensure that the geological bodies on both sides of the fault are not incorrectly connected during interpolation; at the same time, within the fault influence zone (based on the width of the fracture zone), a resistance attenuation zone is set (the resistance coefficient decreases linearly from the fault trace to both sides).

[0081] At the cross-section matching point of the fold axial surface trace, drag gradient marks are set, and the drag coefficients on both sides of the axial surface gradually change according to the inter-wing angle of the fold (the smaller the inter-wing angle, the greater the gradient amplitude), to ensure a smooth transition of the fold shape.

[0082] This embodiment achieves precise association of similar boundaries through feature parameter similarity matching, quantifies resistance transmission using distance weighting, and sets abrupt and gradual change nodes to take into account both the continuity and discontinuity of geological bodies.

[0083] S400, under the action of the resistance constraint network, generates a set of points on the surface of the geological body through an adaptive interpolation algorithm.

[0084] Specifically, the three-dimensional range of the working face is first determined based on the coordinate data of the two roadway profiles and the mining face profile. Then, adaptive mesh density calculation is performed, as follows: A mesh density coefficient is defined. It should be noted that mesh refinement is performed near the profiles: a refinement radius is set with each profile as the center; mesh refinement is further strengthened near key nodes, i.e., a strengthening refinement radius is set with key nodes (fault breaks and fold inflection points) as the center. Three-dimensional mesh nodes are generated within the working face range according to the above density, with node coordinates satisfying either a uniform distribution (unrefined area) or a dense distribution (refined area or strengthened refinement area).

[0085] Next, for each grid node, calculate its Euclidean distance to all boundary feature points. If the Euclidean distance is small (the node is close to the boundary), then associate the node with the drag coefficient and constraint direction (tangent or normal vector) of the corresponding boundary feature point. If the node is located within the fault influence zone, it is marked as a fault influence zone node, and the drag attenuation rule of the fault trace is associated with it. If the node is located in the wing of a fold, it is marked as a fold wing node, and the drag gradient rule of the fold is associated with it.

[0086] This embodiment designs a three-level adaptive density grid based on profile distribution and key node locations, with grid nodes directly associated with resistance constraint information, providing accurate constraint basis for subsequent interpolation. Modeling accuracy is ensured in key areas (near the profile, complex structural areas), while computational load is reduced in non-critical areas. Simultaneously, the association between grid nodes and resistance constraints ensures that the interpolation process is directly controlled by the constraint network.

[0087] Finally, radial basis functions are used as the basic interpolation model, and multiple quadratic functions are selected as the kernel function.

[0088] The closer the Euclidean distance between the grid node and the boundary feature point, the larger the drag coefficient, the larger the correction function value, and the closer the interpolation result is to the coordinates of the feature point; conversely, the farther the Euclidean distance, the smaller the drag coefficient, the smaller the correction function value, and the more the interpolation result depends on the continuity of adjacent nodes.

[0089] The modified interpolation function is obtained based on the modification term and the basic interpolation function.

[0090] Constraint Supplement: The interpolation results in the boundary normal direction are forced to have a displacement of 0 (based on the constraint direction definition) to ensure that the boundary does not drift.

[0091] It should be noted that special handling is required for interpolation of complex geological bodies: For interpolation in fault regions: fault barrier constraints are added to the grid nodes on both sides of the fault during interpolation. That is, nodes on both sides of the fault do not directly participate in each other's interpolation calculations, but are indirectly related only through the resistance constraints of fault trace feature points, avoiding incorrect connections between geological bodies on both sides of the fault. At the same time, the interpolation results of nodes within the fault influence zone are fine-tuned according to the resistance attenuation rule. For nodes in the fold limbs, the interpolation weights are adjusted according to the resistance gradient rule of the fold axial plane trace. That is, nodes closer to the axial plane are more strongly constrained by the axial plane resistance, and the interpolation results are closer to the axial plane morphology; nodes farther from the axial plane are more affected by the continuity weight, ensuring a smooth transition of the fold morphology.

[0092] After interpolation, smoothing is performed using moving least squares smoothing on the interpolated grid node values ​​(Z coordinates). The standard deviation of the interpolation results is calculated, and outlier nodes are removed (interpolation is recalculated). Finally, a geological body surface point set is generated: based on the interpolated grid node Z coordinates, surface point sets (categorized by boundary type) of the coal seam roof and floor, faults, and fold axes are extracted.

[0093] S500 performs three-dimensional morphological fitting on the point set on the surface of the geological body to obtain a three-dimensional model of the coal mining face.

[0094] Specifically, the fitting process for the roof and floor of the coal seam is as follows: Surface reconstruction: The Poisson surface reconstruction algorithm is used to fit the point set of the roof and floor surfaces. The reconstruction depth is set to balance accuracy and smoothness, generating a continuous three-dimensional surface of the roof and floor. Then, thickness calibration is performed, that is, the coal seam thickness at any position is calculated on the fitted roof and floor surface and compared with the previously extracted thickness parameters. If the thickness error is large, the surface control points in the corresponding area are adjusted to ensure that the thickness conforms to the measured data. Finally, the lithology, density, bulk density and other attribute information of the coal seam are mapped to the roof and floor surface and the intermediate coal seam area to form an attributed coal seam model.

[0095] The three-dimensional morphology fitting process of the fault is as follows: fault surface generation: extract the surface point set of the two sides of the fault, and use the triangular patch fitting algorithm to generate the three-dimensional fault surface. Here, the fault trace is the edge line of the surface, and the fault influence zone is the extended area of ​​the surface. Then, mark the fault number, strike, dip, dip angle, displacement, breccia width and other parameters on the fault surface. At the same time, mark the lithological differences between the two sides of the fault on both sides of the surface. Finally, calculate the intersection line between the fault and the top and bottom plates, that is, calculate the intersection line between the fault surface and the top and bottom plate surfaces through Boolean operations to ensure that the intersection line shape is consistent with the fault trace observed in the profile.

[0096] The 3D morphology fitting process of the fold is as follows: Axial surface generation: For the surface point set of the fold axial surface trace, the NURBS surface fitting algorithm (the number of control points = a certain multiple of the number of trace feature points) is used to generate a smooth 3D fold axial surface. Then, based on the interwing angle and tightness of the fold, the curvature of the top and bottom plates of the wing is adjusted. Here, the curvature of the tightly closed fold wing increases, and the curvature of the open fold wing decreases, ensuring that the fold morphology is consistent with the cross-sectional observation results. Finally, the fold integrity is checked, and the symmetry of the two fold wing is calculated. If the asymmetry coefficient is too large, the position and inclination of the axial surface are adjusted to ensure that the fold morphology conforms to geological laws.

[0097] Finally, geological body integration and Boolean operations are performed. First, fitted surfaces such as the roof and floor, faults, and fold axial surfaces are imported, and Boolean operations are performed according to the spatial relationships of the geological bodies. For coal seam areas: the space between the roof and floor surfaces represents the coal seam, extracted using intersection operations; for fault-affected areas: the intersection of the fault surface extension area and the coal seam or surrounding rock represents the fault-affected zone, extracted using union operations; for surrounding rock areas: the space approximately 10m outside the roof and floor represents the surrounding rock, extracted using difference operations. Finally, a complete 3D model is generated: integrating geological bodies such as coal seams, surrounding rock, faults, and folds to form a complete 3D model containing geometric morphology and attribute information, in a triangular patch model format.

[0098] This embodiment employs a dedicated fitting algorithm for different geological bodies, incorporating optimization steps such as thickness calibration and symmetry checks. It also achieves comprehensive mapping of attribute information, resulting in a more detailed reproduction of the three-dimensional morphology of the geological body and more complete attribute information. The model not only reflects the geometric shape but also provides engineering parameters such as lithology and density, providing more comprehensive data support for mining risk assessment and mining volume calculation.

[0099] Furthermore, although the steps of the method in this application are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0100] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A three-dimensional modeling method for coal mining faces based on profile boundary resistance traction, characterized in that, The method includes: Collect geological profile data of the two roadways and the mining face of the coal mining face; Preprocessing and feature extraction were performed on the geological profile data of the two tunnels and the geological profile data of the mining face to obtain the set of boundary feature points; Based on the set of boundary feature points, a cross-section resistance constraint network is constructed; wherein, the resistance constraint network includes cross-section resistance correlation lines and cross-section resistance transmission rules. Under the influence of the resistance constraint network, an adaptive interpolation algorithm is used to generate a set of points on the surface of the geological body. A three-dimensional morphological fitting was performed on the point set on the surface of the geological body to obtain a three-dimensional model of the coal mining face.

2. The three-dimensional modeling method for coal mining faces based on profile boundary resistance traction according to claim 1, characterized in that, Preprocessing includes initial noise filtering and outlier clustering removal; the initial noise filtering includes: Noise filtering was applied to the geological profile data of the two tunnels and the geological profile data of the mining face using Gaussian filtering.

3. The three-dimensional modeling method for coal mining faces based on profile boundary resistance traction according to claim 2, characterized in that, Outlier clustering removal includes: The noise-filtered geological profile data of the two tunnels and the geological profile data of the mining face are converted into two-dimensional feature vectors, with the distance along the profile length and the vertical depth as feature dimensions. The first, second, and third clusters are determined based on two-dimensional feature vectors and the k-means clustering algorithm. The intra-cluster density, dispersion, and distance from other clusters of the first, second, and third clusters are all different. The clusters with the lowest intra-cluster density, the highest dispersion, and the largest distance from other clusters in the first, second, and third clusters are identified as the clusters to be removed. Remove the geological profile data of the two tunnels and the geological profile data of the mining face corresponding to all two-dimensional feature vectors in the cluster to be removed.

4. The three-dimensional modeling method for coal mining faces based on profile boundary resistance traction according to claim 3, characterized in that, The first, second, and third clusters are determined based on two-dimensional feature vectors and the k-means clustering algorithm, including: Randomly select one two-dimensional feature vector from several two-dimensional feature vectors as the candidate first cluster center; Calculate the neighborhood density of the candidate first cluster center; If the neighborhood density is greater than a preset normal neighborhood density threshold, the candidate center is determined to be the first cluster center; Calculate the Euclidean distance from each two-dimensional feature vector except for the first cluster center to the first cluster center; Based on the Euclidean distance from each two-dimensional feature vector other than the first cluster center to the first cluster center, the roulette wheel selection method is used to determine the candidate second cluster center; If the distance from the candidate second cluster center to the first cluster center is greater than the first preset distance threshold, then the candidate second cluster center is determined as the second cluster center; Calculate the Euclidean distance from each two-dimensional feature vector other than the first and second cluster centers to the second cluster center; Based on the Euclidean distance from each two-dimensional feature vector other than the first and second cluster centers to the second cluster center, the roulette wheel selection method is used to determine the candidate third cluster center; If the distance from the candidate third cluster center to the first cluster center is greater than the second preset distance threshold, and the distance to the second cluster center is greater than the third preset distance threshold, then the candidate third cluster center is determined as the third cluster center. Calculate the Euclidean distance from each two-dimensional feature vector (excluding the first, second, and third cluster centers) to the three cluster centers, and assign the data points to the cluster center with the smallest distance to obtain the first, second, and third clusters.

5. The three-dimensional modeling method for coal mining faces based on profile boundary resistance traction according to claim 1, characterized in that, Feature extraction includes: Obtain the corresponding grayscale images, attitude heatmaps, and geological attribute mapping maps of the preprocessed geological profile data of the two roadways and the mining face; among them, the attitude heatmap is an image that converts the dip angle and strike angle of the strata into thermal values; the geological attribute mapping map is an image that converts lithology, density, and wave velocity into multi-channel features; Using the grayscale profile image, occurrence heat map, and geological attribute mapping map as input, feature extraction is performed according to the feature extraction model to obtain the feature parameters of the coal seam roof and floor boundaries, fault traces, and fold axis traces. The feature parameters include the three-dimensional coordinates of the boundary points, strike angle, dip angle, curvature, and specific parameters. The specific parameters are at least one of the following: coal seam thickness, fault displacement, and inter-flange angle.

6. The three-dimensional modeling method for coal mining faces based on profile boundary resistance traction according to claim 1, characterized in that, The feature extraction model employs a composite loss function consisting of weighted cross-entropy and morphological loss. The morphological loss includes top and bottom plate boundary smoothness loss, fault trace continuity loss, and fold axis symmetry loss.

7. The three-dimensional modeling method for coal mining faces based on profile boundary resistance traction according to claim 1, characterized in that, Based on the boundary feature point set, a cross-section resistance constraint network is constructed, including: Determine the tangent and normal directions of each feature point in the boundary feature point set; wherein, the tangent direction is calculated based on the coordinates of adjacent feature points, the normal direction is perpendicular to the tangent direction and points into the geological body, the adjustment range of the feature point along the tangent direction is less than the preset tangent adjustment threshold, and the displacement constraint of the normal direction is 0. Using the geological profiles of the two roadways of the coal mining face as the reference ends and the geological profile of the mining face as the intermediate node, Several similar boundaries are determined by matching the feature parameters of the baseline and intermediate nodes. Connect the feature points of the reference end and the intermediate node corresponding to the same boundary to obtain the cross-sectional resistance correlation line; The resistance coefficient is transmitted according to the distance weighting rule and the adjacent cooperation rule to construct a cross-section resistance constraint network; wherein, the adjacent cooperation rule is triggered when the distance between adjacent intermediate nodes is less than a preset node distance threshold.

8. The three-dimensional modeling method for coal mining faces based on profile boundary resistance traction according to claim 7, characterized in that, The distance-weighted rule meets the following conditions: ； in, Let be the basic resistance coefficient of the i-th intermediate node; The resistance coefficients at the starting points of the geological profiles of the two tunnels; is the resistance coefficient at the end of the geological profile of the two roadways; L is the total length of the coal mining face. is the distance from the i-th intermediate node to the starting point of the geological profile of the two lanes.

Images