Fracture surface reconstruction method and device based on dynamic graph neural network

Abstract

The invention discloses a fissure surface reconstruction method and device based on a dynamic graph neural network. The reconstruction method comprises the following steps: acquiring core parameters of a coal rock fracture source; the method comprises the following steps: preprocessing core parameters of a fracture source by adopting a standardized processing method, constructing a fracture space-time topology analysis model based on a DGCNN graph neural network, and taking fractures as graph nodes; an edge convolution mechanism is utilized, an adjacent graph is constructed through the space-time relation between nodes, neighborhood features are aggregated to update node representation, and a local structure and global topology are captured; crack feature similarity is calculated, and clustering is carried out according to a density threshold value by using a DBSCAN algorithm; and in combination with a crack magnitude optimization clustering result, identifying a main crack surface crack with high expansion potential. Reconstructing a macroscopic crack surface by adopting a B-spline surface fitting algorithm; a control point is selected through a principal component analysis method to determine a crack main direction, a B-spline curve is constructed based on crack point distribution to generate a boundary contour, and a crack surface is accurately fitted by using a B-spline algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of fracture reconstruction, and in particular, to a fracture surface reconstruction method and device based on a dynamic graph neural network. Background Art

[0002] With the increase in the mining depth of coal mine resources, microcracks inside the rock mass gradually expand and form a continuous macroscopic fracture surface, seriously threatening the stability and safety of mine engineering. The formation of continuous fractures is not only a typical multi-scale dynamic evolution phenomenon of the instability and fracture of mine rock masses, but also significantly increases the risk of disaster fractures, causing economic losses and potential safety hazards. Therefore, fracture surface reconstruction is of great significance for revealing the instability evolution mechanism of rock masses and improving the stability of deep mining. In addition, the accurate reconstruction of fracture networks is also crucial in resource development and environmental governance technologies such as hydraulic fracturing and carbon dioxide mineral sequestration. Through fracture surface reconstruction, the spatial distribution, geometric shape, and connection characteristics of fractures can be accurately reproduced, providing a scientific basis for evaluating the effectiveness of artificial fracture systems and optimizing reservoir stimulation designs. Therefore, exploring fracture surface reconstruction methods and their applications in the study of the dynamic evolution of continuous cracks is of great value for disaster prevention and control and energy technology development.

[0003] The formation and penetration of fracture surfaces are complex "black box" evolution processes, and it is difficult for traditional methods to reveal their mechanisms. As a non-destructive detection technology, microseismic monitoring provides an effective means to open this "black box" by capturing the elastic wave signals associated with coal and rock under loading. Through microseismic inversion, not only can the spatial position of the fracture source be accurately located, but also the spatio-temporal dynamic process and fracture mechanism of crack propagation can be quantitatively deduced, providing scientific support for studying the dynamic evolution law, penetration characteristics and induced disaster mechanism of internal cracks in rock masses. Patent CN114966849A provides a method for characterizing rock mass fractures based on microseismic or acoustic emission and focal mechanism constraints, and realizes fracture characterization by clustering and fitting fracture surfaces from microseismic point cloud data. Patent CN115077437A discloses a method for characterizing the morphology of rock hydraulic fracturing fractures based on acoustic emission location constraints, generating point clouds through acoustic emission location, and segmenting and optimizing small planes to determine fracture orientation and geometric parameters. Patent CN118884524A proposes a method for constructing a fracture network model and identifying seepage channels based on microseismic monitoring, using microseismic moment tensors and source parameters to construct a seepage model and identify the main channels with the minimum flow-through time. Although these fracture characterization methods can reveal the spatial distribution characteristics of fractures to a certain extent and achieve three-dimensional description from the perspective of point clouds, most of them rely on distance or energy clustering of point clouds, do not fully consider the complex spatio-temporal topological correlations between fracture points, and lack comprehensive analysis of the multi-dimensional characteristics of fractures, restricting the in-depth understanding and characterization of the global characteristics of fracture networks. In this context, deep learning methods, especially graph neural network technology, have become the key to breaking through traditional methods. Graph neural networks have significant advantages in capturing spatio-temporal correlations between nodes, modeling physical connectivity between nodes, and extremely high flexibility and efficiency in dealing with complex topological relationships, enabling them to efficiently process complex fracture point cloud data and providing a new idea and tool for fracture surface reconstruction. Therefore, a method for intelligently reconstructing fracture surfaces using a graph neural network model is proposed to achieve full-automatic inversion and reconstruction of fracture surfaces in the spatio-temporal dimension, comprehensively describe the complex spatio-temporal topological relationships and penetration mechanisms between fracture points, and provide strong support for deeply understanding the multi-scale dynamic evolution process of rock mass instability and fracture. Summary of the Invention

[0004] In response to the problems and requirements raised above, this solution proposes a method and device for reconstructing fracture surfaces based on a dynamic graph neural network, which can achieve the above technical objectives and bring many other technical effects due to the following technical features.

[0005] An object of the present invention is to propose a method for reconstructing fracture surfaces based on a dynamic graph neural network, including the following steps:

[0006] S10: Collect the original acoustic wave data through a microseismic monitoring system, perform positioning calculations and focal mechanism inversion on the original acoustic wave data to obtain the core parameters of the coal-rock fracture source directly related to the fracture; among them, the core parameters of the coal-rock fracture source include: fracture spatial coordinates, fracture volume, fracture energy, fracture spatial orientation, and fracture magnitude;

[0007] S20: Preprocess the core parameters of the fracture source using a standardization processing method, construct a fracture spatio-temporal topology analysis model based on the DGCNN graph neural network, and use fractures as graph nodes; utilize the edge convolution mechanism to construct an adjacency graph through the spatio-temporal relationship between nodes, and aggregate neighborhood features to update the node representation, capturing local structures and global topologies;

[0008] S30: Calculate the fracture feature similarity based on the Euclidean distance, use the DBSCAN algorithm to cluster according to the density threshold, group fracture points with similar features into the same cluster, each cluster represents a fracture surface, and assign a unique number; and optimize the clustering results in combination with the fracture magnitude to identify the main fracture surface fractures with high expansion potential;

[0009] S40: Reconstruct the main fracture surface using the B-spline surface fitting algorithm to generate a smooth and continuous macroscopic fracture surface; select control points through the principal component analysis method to determine the main direction of the crack, construct a B-spline curve based on the fracture point distribution to generate the boundary contour, triangulate the outer contour line and divide the grid, combine the fracture points and the triangular grid, and use the B-spline algorithm to accurately fit the fracture surface.

[0010] In an example of the present invention, in the step S10, perform positioning calculations and focal mechanism inversion on the original acoustic wave data to obtain the core parameters of the focal mechanism directly related to the fracture, including the following steps:

[0011] S11: Use the arrival time data of the first waves of acoustic waves at sensors in different positions of the microseismic monitoring system, and use the simplex positioning robust algorithm to perform spatial positioning calculations on the acoustic emission source to obtain the fracture spatial coordinates;

[0012] S12: Use the arrival time data of the first waves of acoustic waves at sensors in different positions of the microseismic monitoring system, and use the simplex positioning robust algorithm to perform spatial positioning calculations on the acoustic emission source;

[0013] S13: Based on the constraint conditions of the tensile-shear fracture source model, solve the 6 components M of the source fracture moment tensor from the first wave amplitude data collected by sensors in different positions pq ;

[0014] S14: According to the calculated moment tensor components, quantitatively invert and calculate the core parameters of the coal-rock fracture source, where the core parameters of the coal-rock fracture source include fracture spatial coordinates, fracture volume, fracture energy, and fracture spatial orientation.

[0015] In an example of the present invention, in the step S14, the specific calculation formulas for the core parameters of the coal-rock fracture source are as follows:

[0016] Fracture volume:

[0017]

[0018] In the formula, ΔV is the fracture volume of the fracture point, M 1 、M 3 are the eigenvalues of the moment tensor M pq , and μ is the Lame constant;

[0019] Fracture spatial orientation:

[0020]

[0021] n = (cosα, 0, ±sinα)

[0022]

[0023] In the formula, α is the angle between the movement direction and the normal direction of the source fracture surface, M 1 、M 2 、M 3 are the eigenvalues of the moment tensor M pq , μ and λ are the Lame constants, n is the spatial orientation, and b is the movement direction;

[0024] Fracture energy:

[0025]

[0026] In the formula: σ t is the tensile strength, σ s is the shear strength, ΔA is the fracture area of the fracture point, and σ is the material strength;

[0027] Fracture magnitude:

[0028]

[0029] In the formula, M 1 、M 2 、M 3 are the eigenvalues of the moment tensor M pq .

[0030] In an example of the present invention, in the step S20, a normalization processing method is used to preprocess the core parameters of the fracture source; specifically, it includes the following:

[0031] Assume that the preprocessed core parameters of the fracture source are x 1 , x 2 , …, x n-1 , xn , it is mapped to the interval [0, 1] through the min-max normalization method, and its normalization formula is as follows:

[0032]

[0033] In the formula, is the minimum value of the preprocessing parameter, is the maximum value of the preprocessing parameter, y i ∈ [0, 1].

[0034] In an example of the present invention, in the step S20, the crack spatio-temporal topology analysis model includes a space transformation module, an edge convolution module, a max pooling layer, and a fully connected layer.

[0035] The space transformation module is configured to normalize and align the input point cloud data, and obtain an affine transformation matrix through learning, which is used to convert the input point cloud data into a canonical coordinate system; this helps to eliminate the rotation and translation differences of the point cloud data, thereby improving the robustness and generalization ability of the network.

[0036] The edge convolution module is configured to perform convolution operations for feature extraction and modeling local structure information on the point cloud data, and dynamically learn and extract local structure features.

[0037] The max pooling layer is configured to aggregate the features of the crack points and extract the most representative local information.

[0038] The fully connected layer is configured to map the global features to a higher-dimensional space in order to capture complex non-linear patterns and potential relationships in the data.

[0039] In an example of the present invention, convolution operations for feature extraction and modeling local structure information are performed on the point cloud data, and local structure features are dynamically learned and extracted, which specifically include the following steps:

[0040] (a) Input of point cloud data: Assume that the input point cloud data contains n crack points, then the point cloud data X is expressed as:

[0041] X = {x 1 , x 2 ,....x n} ∈ R D

[0042] In the formula: x i = {px i , py i , pz i , t i , nx i , ny i , nzi , V i , E i} is the feature of the i-th fracture point; D is the feature dimension; {px i , py i , pz i} is the spatial coordinate of the i-th fracture point; t i is the timestamp of the i-th fracture point; {nx i , ny i , nz i} is the spatial orientation of the i-th fracture point; V i is the fracture volume of the i-th fracture point; E i is the released energy of the i-th fracture point;

[0043] (b) Establish an adjacency relationship: The adjacency relationship is used to establish the graph structure between point clouds. Each fracture point serves as a node of the graph, and the adjacency relationship between nodes needs to be dynamically constructed according to predefined rules; among them, it is judged whether two fracture points are connected based on temporal adjacency, spatial topological relationship, spatial orientation similarity, and energy similarity;

[0044] (c) Construct a graph structure: Define a directed graph G for representing the local point cloud structure. G is composed of vertices V and directed edges E, and its expression is:

[0045] G = (V, E), E ∈ V × V

[0046] Where: V i = {x 1 , x 2 ,....x n} is the set of all located points, represents the set of directed edges of x i and its k adjacent vertices;

[0047] (d) Edge feature extraction: Calculate by obtaining local features with the difference between the located point and its adjacent points, and use a multi-layer perceptron f Θ to extract edge features. Assume that the edge feature between the located point x i and its j-th neighboring point is:

[0048]

[0049] Where: f Θ is a series of non-linear functions parameterized by the set of learnable parameters Θ;

[0050] (e) Feature aggregation: After extracting the edge features of each center point, first perform a linear transformation through a fully connected layer, and then perform a normalization process; then aggregate the features of the center point with the features of the points within its local neighborhood through an aggregation function to obtain the aggregated features, add the input center point features to the aggregated center point features to achieve residual connection; finally, pass through the ReLU activation function to achieve non-linear mapping to obtain the output of the edge convolution network.

[0051] In an example of the present invention, in step (b) of establishing the adjacency relationship, determining whether two crack points establish a connection needs to simultaneously satisfy the following conditions:

[0052] Temporal adjacency: The difference in time between two positioning points should be less than a time threshold, indicating that the times of their occurrences are close and they may belong to the same crack propagation process:

[0053] |t i -t j |≤Δt

[0054] Where: t i , t j are the rupture times of crack points i and j respectively, and Δt is a predefined time threshold;

[0055] Spatial topological relationship: By calculating the geometric relationship between crack points, three topological relationships are defined: intersection, adjacency, and separation; only when two crack points intersect or are adjacent can an adjacency relationship be established, indicating that they are close enough in space and can penetrate and belong to the same crack;

[0056] Among them, calculate the spatial distance d ij :

[0057]

[0058] Where: (x i , y i , z i ), (x j , y j , z j ) are the positioning spatial coordinates of crack points i and j respectively;

[0059] Spatial orientation similarity: The spatial orientation vectors of two positioning points need to be similar within a certain error range, then it is considered that they belong to the same crack, and the similarity of the directions is judged by calculating the included angle between the direction vectors. Among them, the expression of the included angle between the direction vectors is:

[0060]

[0061] Where, n i=(nx i , ny i , nz i ) and n j =(nx j , ny j , nz j ) are the spatial orientation vectors of two fracture points respectively, and Δθ is the predefined angular error range;

[0062] Energy similarity: The energy difference between two positioning points should be less than a certain threshold, indicating that they have similar energy release patterns and can be physically connected or penetrated. The judgment expression is:

[0063] |E i - E j | ≤ ΔE

[0064] In the formula: E i , E j are the released energy values of fracture points i and j respectively, and ΔE is the predefined exponential threshold.

[0065] In an example of the present invention, in the step S30, clustering is performed according to the density threshold by using the DBSCAN algorithm, including the following steps:

[0066] S31: Feature vector reading and Euclidean distance calculation: Read the high-dimensional feature vectors of each fracture point obtained through the fracture spatio-temporal topology analysis model, and calculate the Euclidean distance between the fracture points according to the high-dimensional feature vectors to measure the similarity between the fracture points; among them, for each pair of fracture points P i =(x 1 , x 2 , x 3 ....x n ) and P j =(y 1 , y 2 , y 3 ....y n ), the Euclidean distance calculation formula is as follows:

[0067]

[0068] In the formula: n is the dimension of the feature vector, x k and y k are the values of the k-th dimension in the high-dimensional feature vectors of two fracture points respectively;

[0069] S32: Determine the neighborhood distance and the minimum number of points: Select an appropriate neighborhood distance ε and the minimum number of points MinPts; Based on the Euclidean distance and the neighborhood distance ∈, calculate the neighborhood of each crack point. If the number of points contained in the neighborhood of a crack point is greater than or equal to MinPts, then this crack point is a core point; Otherwise, it is a border point or a noise point;

[0070] S33: Cluster expansion and core point processing: Select an unlabeled core point as the initial point, label it as the starting point of a new cluster, and assign a cluster number to this point; Expand the neighborhood where this core point is located, and assign all points in the neighborhood to the same cluster; For the newly added points in the expanded cluster, check whether it is a core point; If it is a core point, continue to expand the cluster; If it is a border point, keep its cluster label unchanged; Continue to select new points from the unlabeled core points for cluster expansion until all core points have been processed;

[0071] S34: Cluster numbering and clustering result checking: According to the result of the DBSCAN clustering in step S33, assign a unique crack surface number to each cluster, and each cluster represents an independent crack surface; If a certain crack point belongs to the neighborhoods of multiple core points at the same time, then this point will be assigned the crack surface number of the first discovered cluster; Then check the clustering result to ensure that each cluster is assigned a unique crack surface number, and the number of noise points is -1; If it is found that the cluster number assignment is incorrect or missing, make necessary adjustments to ensure the accuracy of the final crack surface number.

[0072] S35: Optimization of the clustering result based on magnitude: After obtaining the preliminary clustering result, optimize the clustering result by combining the magnitude of each crack point. By visualizing the magnitude distribution of the crack points within the crack surface, identify the crack points with larger magnitudes, that is, those belonging to the main crack surface; Points with small magnitudes indicate secondary crack surfaces or crack surfaces with less expansion potential.

[0073] In an example of the present invention, in the step S40, the B-spline surface fitting algorithm is used to reconstruct the crack points of the main crack surface to generate a smooth and continuous macroscopic crack surface, including the following steps:

[0074] S41: From the result of the clustering analysis, read all the valid crack points that have been classified into the same crack surface; Among them, each crack point includes the crack spatial coordinates, the crack volume, the crack spatial orientation, the crack energy, and the crack magnitude;

[0075] S42: Combine the energy and the normal direction to comprehensively describe the spatial characteristics and the fracture influence of the crack points;

[0076] Assume that the normal direction vector of a certain crack point is n = (n x , n y , n z) The calculation formula for the modulus length of this vector is as follows:

[0077]

[0078] Normalize the normal vector to ensure that the normalized normal vector only represents the direction. The calculation formula for its normalization process is as follows:

[0079]

[0080] For each fracture point, calculate the comprehensive feature weight according to its fracture energy and normal direction; assume that the feature weight W i of the fracture point i is a combination of fracture energy and normal direction. Then the calculation formula for the comprehensive feature weight of each fracture point is as follows:

[0081]

[0082] In the formula, is the modulus length of the normalized normal direction, usually 1; E i is the fracture energy of the i-th fracture point; α and β are weight coefficients;

[0083] S43: Based on the feature weights and three-dimensional spatial position information of the fracture points in the same cluster, extract the fracture points with higher feature weights through principal component analysis as the key control points for fracture surface reconstruction, so as to clarify the main direction of the fracture surface;

[0084] S44: After determining the main direction of the crack surface, use the B-spline curve fitting method to fit the boundaries of the fracture points in the same cluster to form a closed-loop curve as the outer contour line of the final crack surface;

[0085] S45: Triangulate the outer contour line of the already formed crack surface and perform triangular mesh division on the interior of the final crack surface;

[0086] S46: Read all the fracture points in the same cluster, and combine with the already established triangular mesh to finally fit the crack surface to be generated to form the final macroscopic crack surface of the fracture points in this cluster.

[0087] Another object of the present invention is to propose a fracture surface intelligent reconstruction device based on a dynamic graph neural network, including:

[0088] A parameter acquisition and obtaining module, configured to collect original acoustic wave data through a microseismic monitoring system, perform positioning calculation and seismic source mechanism inversion on the original acoustic wave data, and obtain the core parameters of the coal and rock fracture source directly related to the fracture;

[0089] The model construction module is configured to preprocess the core parameters of the fracture source using a standardization processing method, construct a crack spatio-temporal topology analysis model based on the DGCNN graph neural network, and use cracks as graph nodes; utilize the edge convolution mechanism to construct an adjacency graph through the spatio-temporal relationship between nodes, aggregate neighborhood features to update node representations, and capture local structures and global topologies; the crack surface clustering module is configured to calculate the similarity of crack features based on the Euclidean distance, use the DBSCAN algorithm to cluster according to the density threshold, group crack points with similar features into the same cluster, each cluster representing a crack surface and assigned a unique number; and optimize the clustering results in combination with the crack magnitude to identify the main crack surface cracks with high expansion potential;

[0090] The crack surface reconstruction module is configured to reconstruct the main crack surface using the B-spline surface fitting algorithm to generate a smooth and continuous macroscopic crack surface; select control points through the principal component analysis method to determine the main crack direction, construct a B-spline curve based on the distribution of crack points to generate a boundary contour, triangulate the outer contour line and divide the grid, and combine the crack points and the triangular grid to accurately fit the crack surface using the B-spline algorithm.

[0091] The present invention has the following beneficial effects compared with the prior art:

[0092] 1. The present invention obtains the core parameters of the source mechanism through microseismic monitoring inversion, which can directly reflect the source characteristics of crack points, and has high reliability and accuracy. Compared with the traditional characterization method that only relies on spatial positioning points, the core parameters provided by the source mechanism inversion as data input can more accurately capture the spatio-temporal correlation and physical connectivity between cracks, significantly improving the scientificity and reliability of the intelligent reconstruction of crack surfaces. At the same time, it is closer to the true evolution law of internal cracks in rock masses, laying a solid data foundation for analyzing the crack propagation and penetration mechanism.

[0093] 2. The present invention constructs a crack spatio-temporal topology analysis model based on a dynamic graph neural network, captures the spatio-temporal correlation and global topology characteristics of crack points through edge convolution, and uses the adjacency relationship to deeply explore the spatio-temporal correlation, physical connectivity and topological structure between crack points, breaking through the limitations of traditional methods in dealing with complex crack networks, significantly improving the accuracy of crack network analysis and topological characterization ability, and providing strong technical support for the intelligent reconstruction of crack surfaces.

[0094] 3. To overcome the deficiencies of traditional crack characterization methods in spatio-temporal topology correlation and penetration mechanism analysis, the invention improves the adjacency relationship in the graph neural network, further integrates the spatial orientation and energy dissipation characteristics of crack points on the basis of considering the relative relationship between the spatial position and time series of crack points, deeply explores the physical connectivity and topological structure between crack points, and enables it to more effectively capture and analyze the complex spatio-temporal topology relationship between crack points.

[0095] 4. Through the combination of graph neural network modeling, DBSCAN clustering, and B-spline fitting, the present invention can reconstruct the geometric morphology and dynamic evolution characteristics of fracture surfaces with high precision, intuitively display the expansion process, temporal evolution, and spatial distribution of macroscopic fracture surfaces, greatly improve the accuracy and reliability of fracture surface inversion, provide advanced technical support for in-depth analysis of the rock mass fracture evolution mechanism, and at the same time provide a more scientific basis for the stability assessment and disaster warning of rock mass engineering.

[0096] In the following, the optimal embodiments of implementing the present invention will be described in more detail with reference to the accompanying drawings, so as to facilitate the understanding of the features and advantages of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly introduced below. Among them, the accompanying drawings are only used to show some embodiments of the present invention, rather than limiting all embodiments of the present invention thereto.

[0098] Figure 1 FIG. is a flowchart of a fracture surface reconstruction method based on a dynamic graph neural network according to an embodiment of the present invention;

[0099] Figure 2 FIG. is a schematic diagram of the principle of DGCNN according to an embodiment of the present invention;

[0100] Figure 3 FIG. is a schematic diagram of a convolutional structure according to an embodiment of the present invention;

[0101] Figure 4 FIG. is a flowchart of edge convolution according to an embodiment of the present invention;

[0102] Figure 5 FIG. is a flowchart of DBSCAN clustering according to an embodiment of the present invention.

[0103] Figure 6 FIG. is a flowchart of fitting a fracture surface with a B-spline surface according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0104] In order to make the objectives, technical solutions, and advantages of the technical solutions of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the specific embodiments of the present invention. The same reference numerals in the drawings represent the same components. It should be noted that the described embodiments are some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0105] Unless otherwise defined, the technical terms or scientific terms used herein shall have the ordinary meanings as understood by those of ordinary skill in the art to which the present invention pertains. The terms "first", "second" and similar terms used in the specification and claims of this patent application for invention do not denote any order, quantity or importance, but are only used to distinguish different components. Similarly, terms such as "a" or "an" do not necessarily denote a limitation in quantity. Terms such as "comprising" or "including" mean that the elements or items appearing before this term cover the elements or items listed after this term and their equivalents, without excluding other elements or items. Terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. Terms such as "upper", "lower", "left" and "right" are only used to indicate relative position relationships, and when the absolute position of the object being described changes, the relative position relationships may also change accordingly.

[0106] A method for reconstructing a fracture surface based on a dynamic graph neural network according to a first aspect of the present invention, as Figure 1 shown, includes the following steps:

[0107] S10: Collect the original acoustic wave data through a microseismic monitoring system, perform positioning calculations and source mechanism inversions on the original acoustic wave data, and obtain the core parameters of the coal and rock fracture source directly related to the fracture; wherein, the core parameters of the coal and rock fracture source include: fracture spatial coordinates, fracture volume, fracture energy, fracture spatial orientation, and fracture magnitude.

[0108] S20: Preprocess the core parameters of the fracture source using a standardization processing method, construct a fracture spatio-temporal topology analysis model based on the DGCNN graph neural network, take the fractures as graph nodes, and the node features include time, spatial coordinates, volume, energy, and spatial orientation; use the edge convolution mechanism to construct an adjacency graph through the spatio-temporal relationship between nodes, and aggregate the neighborhood features to update the node representation, capturing the local structure and global topology; wherein, to ensure the quality and applicability of the input data, the data obtained from the inversion calculation is preprocessed to provide standardized data for the input of the subsequent graph neural network model, avoiding interference from data inconsistency to model training, thereby improving the accuracy and generalization ability of the model; wherein, the parameters are optimized by minimizing the loss function, and the model outputs a high-dimensional feature vector, comprehensively reflecting the spatio-temporal correlation, physical connectivity, and topological structure;

[0109] S30: Calculate the similarity of fracture features based on the Euclidean distance, use the DBSCAN algorithm to cluster according to the density threshold, group the fracture points with similar features into the same cluster, each cluster represents a crack surface, and assign a unique number, and the noise points are excluded to reduce interference; and optimize the clustering result in combination with the fracture magnitude to identify the main crack surface fractures with high expansion potential;

[0110] S40: Reconstruct the main crack surface using the B-spline surface fitting algorithm to generate a smooth and continuous macroscopic crack surface; determine the main crack direction by selecting control points through the principal component analysis method, construct a B-spline curve based on the distribution of fissure points to generate the boundary contour, triangulate the outer contour line and divide the grid, and combine the fissure points and triangular grid to accurately fit the fissure surface using the B-spline algorithm.

[0111] This method obtains the core parameters of the focal mechanism through microseismic monitoring inversion, which can directly reflect the source characteristics of fissure points, and has high reliability and accuracy. Compared with the traditional characterization method that only relies on spatial positioning points, the core parameters provided by the focal mechanism inversion as data input can more accurately capture the spatio-temporal correlation and physical connectivity between fissures, significantly improving the scientificity and reliability of the intelligent reconstruction of the fissure surface. At the same time, it is closer to the true evolution law of internal fissures in rock masses, laying a solid data foundation for analyzing the fissure propagation and penetration mechanism.

[0112] This method constructs a spatio-temporal topology analysis model of fissures based on a dynamic graph neural network, captures the spatio-temporal correlation and global topology characteristics of fissure points through edge convolution, and uses the adjacency relationship to deeply explore the spatio-temporal correlation, physical connectivity, and topological structure between fissure points, breaking through the limitations of traditional methods in dealing with complex fissure networks, significantly improving the accuracy and topological characterization ability of fissure network analysis, and providing strong technical support for the intelligent reconstruction of the fissure surface.

[0113] To overcome the deficiencies of traditional fissure characterization methods in spatio-temporal topology correlation and penetration mechanism analysis, this method improves the adjacency relationship in the graph neural network. On the basis of considering the relative relationship between the spatial position and time series of fissure points, it further integrates the spatial orientation and energy dissipation characteristics of fissure points, deeply explores the physical connectivity and topological structure between fissure points, and enables it to more effectively capture and analyze the complex spatio-temporal topology relationship between fissure points.

[0114] This method combines graph neural network modeling, DBSCAN clustering, and B-spline fitting, which can accurately reconstruct the geometric morphology and dynamic evolution characteristics of the fissure surface, intuitively display the expansion process, temporal evolution, and spatial distribution of the macroscopic fissure surface, greatly improving the accuracy and reliability of fissure surface inversion, providing advanced technical support for in-depth analysis of the rock mass fracture evolution mechanism, and at the same time providing a more scientific basis for rock mass engineering stability assessment and disaster warning.

[0115] In an example of the present invention, in the step S10, perform positioning calculation and focal mechanism inversion on the original acoustic wave data to obtain the core parameters of the focal mechanism directly related to the fracture, including the following steps:

[0116] S11: Post-process the original acoustic wave data collected by the microseismic monitoring system, and use the Akaike information criterion method to extract the first arrival time and amplitude information;

[0117] S12: Use the first arrival time data of acoustic waves at sensors in different positions of the microseismic monitoring system, and use the simplex location robust algorithm to perform spatial location calculation on the acoustic emission source to obtain the fracture space coordinates. Among them, the location control equation is as follows:

[0118]

[0119] In the formula, x(x,y,z) and x i (x i ,y i ,z i ) are the spatial coordinates of the acoustic emission source and the i-th sensor respectively, c p is the P-wave velocity, t i is the P-wave arrival time at the i-th sensor, and t is the earthquake occurrence time of the acoustic emission source;

[0120] S13: Based on the constraint conditions of the tensile-shear fracture source model, solve the 6 components M of the source fracture moment tensor from the first arrival amplitude data of acoustic waves collected by sensors in different positions pq ;

[0121] S14: According to the calculated moment tensor components, quantitatively invert and calculate the core parameters of the coal and rock fracture source. Among them, the core parameters of the coal and rock fracture source include fracture volume, fracture energy, fracture spatial orientation, and fracture magnitude.

[0122] In an example of the present invention, in the step S14, the specific calculation formulas of the core parameters of the coal and rock fracture source are as follows:

[0123] Fracture volume:

[0124]

[0125] In the formula, ΔV is the fracture volume of the fracture point, M 1 , M 3 are the eigenvalues of the moment tensor M pq , and μ is the Lame constant;

[0126] Fracture spatial orientation:

[0127]

[0128] n = (cosα, 0, ±sinα)

[0129]

[0130] In the formula, α is the angle between the movement direction and the normal direction of the source fracture surface, M1 , M 2 , M 3 is the moment tensor M pq 's eigenvalue, μ and λ are Lame constants, n is the spatial orientation, and b is the direction of motion;

[0131] Fracture energy:

[0132]

[0133] Where: σ t is the tensile strength, σ s is the shear strength, ΔA is the fracture area of the fracture point, and σ is the material strength;

[0134] Fracture magnitude:

[0135]

[0136] Where M 1 , M 2 , M 3 are the eigenvalues of the moment tensor M pq .

[0137] In an example of the present invention, in the step S20, a standardization processing method is used to preprocess the core parameters of the fracture source to ensure that the parameter values are within the same scale range and avoid interference of parameters with different dimensions on subsequent analysis; specifically, it includes the following:

[0138] Assume that the core parameters of the preprocessed fracture source are x 1 , x 2 , …, x n-1 , x n , and they are mapped to the interval [0, 1] through the min-max normalization method. Its standardization formula is as follows:

[0139]

[0140] Where is the minimum value of the preprocessing parameter, is the maximum value of the preprocessing parameter, and y i ∈[0, 1].

[0141] In an example of the present invention, in the step S20, as Figure 2 and Figure 3 shown, the fracture spatio-temporal topology analysis model includes a spatial transformation module, an edge convolution module, a max pooling layer, and a fully connected layer,

[0142] The spatial transformation module is configured to normalize and align the input point cloud data, and through learning, an affine transformation matrix is obtained for converting the input point cloud data into a canonical coordinate system; this helps to eliminate the rotational and translational differences in the point cloud data, thereby improving the robustness and generalization ability of the network.

[0143] The edge convolution module is configured to perform convolution operations for feature extraction and modeling local structure information on the point cloud data, dynamically learning and extracting local structure features. Each point in the point cloud data is regarded as a vertex of a graph. For each point, a set of points related to this point can be determined by defining the adjacency relationship, thereby obtaining a directed graph with weighted edges. Then, the features in these local neighborhoods are aggregated to obtain the local feature representation of the point. To update the feature representation of each point, a multi-layer perceptron or a dot product operation between the convolution kernel and the feature can be used. Through these operations, a higher-level feature representation of each point can be extracted. Finally, edge convolution summarizes the features of all points through max pooling to obtain the global feature representation of the entire point cloud data.

[0144] The max pooling layer, as a way of global pooling, is configured to aggregate the features of the crack points and extract the most representative local information; within the neighborhood of each crack point, max pooling selects the maximum value, thereby focusing on the most significant features, which helps to improve the robustness and stability of the model. Specifically, max pooling endows the model with translational invariance by reducing the impact of spatial transformation on the model, that is, regardless of how the position or spatial configuration of the crack points changes, the pooled features remain consistent. This property enables DGCNN to effectively process unordered acoustic emission crack point cloud data, identify common patterns and structural features of crack propagation, and then capture changes in spatio-temporal relationships. Finally, the global features aggregated by the max pooling layer will be passed as input to the fully connected layer to provide a high-level feature representation for further crack clustering and path analysis.

[0145] The fully connected layer is configured to further process and optimize the pooled global features. In DGCNN, the max pooling layer aggregates the local features of the point cloud into global features, while the fully connected layer is responsible for mapping these global features to a higher-dimensional space to capture complex non-linear patterns and potential relationships in the data. The role of the fully connected layer is to enhance the expressive ability of the features through successive linear transformations and non-linear activations. After being processed by multiple fully connected layers, the network will output a high-dimensional feature vector, which synthesizes the global information and non-linear relationships of the input point cloud data and can effectively represent the features of the entire point cloud.

[0146] In an example of the present invention, as Figure 4As shown, convolution operations for feature extraction and modeling local structural information are performed on the point cloud data to dynamically learn and extract local structural features, which specifically include the following steps:

[0147] (a) Point cloud data input: Assume that the input point cloud data contains n fracture points, then the point cloud data X is expressed as:

[0148] X = {x 1 , x 2 ,....x n} ∈ R D

[0149] Where: x i = {px i , py i , pz i , t i , nx i , ny i , nz i , V i , E i} is the feature of the i-th fracture point; D is the feature dimension; {px i , py i , pz i} is the spatial coordinate of the i-th fracture point; t i is the timestamp of the i-th fracture point; {nx i , ny i , nz i} is the spatial orientation of the i-th fracture point; V i is the fracture volume of the i-th fracture point; E i is the released energy of the i-th fracture point;

[0150] (b) Establishing adjacency relationships: Adjacency relationships are used to establish the graph structure between point clouds. Each fracture point serves as a node in the graph, and the adjacency relationships (edges) between nodes need to be dynamically constructed according to predefined rules. Among them, two fracture points are judged whether to establish a connection based on temporal adjacency, spatial topological relationships (including spatial distance and volume), spatial orientation similarity, and energy similarity. That is to say, if all the above four conditions are met, the next step is executed; otherwise, the next fracture point is judged;

[0151] (c) Constructing the graph structure: Define a directed graph G for representing the local point cloud structure. G is composed of vertices V and directed edges E, and its expression is:

[0152] G = (V, E), E ∈ V × V

[0153] Where: V i = {x 1 , x 2,....x n}, the set of all positioning points represents x i the set of directed edges of its adjacent k vertices

[0154] (d) Edge feature extraction: Calculate in the way of obtaining local features by using the positioning points and the differences between the positioning points and their adjacent points, and use the multi-layer perceptron f Θ to extract edge features. Assume the positioning point x i and its j-th neighboring point The edge feature between them is as follows:

[0155]

[0156] In the formula: f Θ is a series of non-linear functions parameterized by the set of learnable parameters Θ

[0157] (e) Feature aggregation: After extracting the edge features of each center point, first perform a linear transformation through a fully connected layer, and then perform a normalization process to ensure that the mean of the data is 0 and the variance is 1; then, aggregate the features of the center point and the features of the points in its local neighborhood through an aggregation function to obtain the aggregated features. To prevent the phenomenon of gradient disappearance during the training process, add the input center point features and the aggregated center point features to achieve residual connection; finally, pass through the ReLU activation function to achieve non-linear mapping to obtain the output of the edge convolution network.

[0158] In an example of the present invention, in step (b) for establishing an adjacency relationship, determining whether two crack points establish a connection needs to simultaneously meet the following conditions:

[0159] Temporal adjacency: The difference in time between two positioning points should be less than a time threshold, indicating that the times when they occur are close and they may belong to the same crack propagation process:

[0160] |t i -t j | ≤ Δt

[0161] In the formula: t i , t j are the rupture times of crack points i and j respectively, and Δt is a predefined time threshold;

[0162] Spatial topological relationship: By calculating the geometric relationship between crack points, three topological relationships are defined: intersection, adjacency, and separation; only when two crack points intersect or are adjacent can an adjacency relationship be established, indicating that they are close enough in space and can penetrate and belong to the same crack;

[0163] Among them, calculate the spatial distance d between crack point i and other crack points j ij :

[0164]

[0165] In the formula: (x i , y i , z i ), (x j , y j , z j ) are the positioning spatial coordinates of crack points i and j respectively;

[0166] Considering the calculation error of crack geometric parameters, increase the radius of the crack by 5% and decrease it by 5% respectively as the inner and outer diameters of the crack, and establish the spatial topological relationship between cracks:

[0167] Suppose the inner and outer diameters of crack points x i and x j are r i,in , r j,in and r i,out , r j,ou ;

[0168] Intersection: If the inner diameters of two crack points xi and xj intersect, that is, d ij ≤ (r i,in + r j,in ), it means that the crack points have a direct overlap in space, may belong to the same crack, or at least have a physical connection with each other, and an adjacency relationship can be established;

[0169] Adjacency: If the outer diameters of two crack points xi and xj intersect but the inner diameters do not intersect, that is, d ij ≤ (r i,out + r j,out ) and d ij > (r i,in + r j,in ), it means that the crack points partially overlap, may be on different propagation paths, and an adjacency relationship can be established, but the connection degree is weak;

[0170] Separation: If there is no intersection between the inner and outer diameters of two crack points x i and x j , and the distance between them is greater than the sum of the outer diameters, that is, d ij > (r i,out + r j,out ), it means that there is no physical connection between the two crack points and there is no adjacency relationship;

[0171] Spatial orientation similarity: The spatial orientation vectors of two positioning points need to be similar within a certain error range, and they are considered to belong to the same crack. The similarity of the directions is judged by calculating the angle between the direction vectors. The expression for the angle between the direction vectors is as follows:

[0172]

[0173] In the formula, n i =(nx i , ny i , nz i ) and n j =(nx j , ny j , nz j ) are the spatial orientation vectors of two fracture points respectively, and Δθ is the predefined angle error range;

[0174] Energy similarity: The energy difference between two positioning points should be less than a certain threshold, indicating that they have a similar energy release pattern and can be physically connected or penetrated. The judgment expression is:

[0175] |E i - E j | ≤ ΔE

[0176] In the formula: E i , E j are the released energy values of fracture points i and j respectively, and ΔE is the predefined exponential threshold.

[0177] In an example of the present invention, in the step S30, as Figure 5 shown, clustering is performed using the DBSCAN algorithm according to the density threshold, including the following steps:

[0178] S31: Feature vector reading and Euclidean distance calculation: Read the high-dimensional feature vectors of each fracture point obtained through the fracture spatio-temporal topology analysis model, and calculate the Euclidean distance between the fracture points according to these high-dimensional feature vectors to measure the similarity between the fracture points; among them, for each pair of fracture points P i =(x 1 , x 2 , x 3 ....x n ) and P j =(y 1 , y 2 , y 3 ....y n ), the Euclidean distance calculation formula is as follows:

[0179]

[0180] where: n is the dimension of the feature vector, x k and y k are respectively the values of the k-th dimension in the high-dimensional feature vectors of two crack points;

[0181] S32: Determine the neighborhood distance and the minimum number of points: Select an appropriate neighborhood distance ε and the minimum number of points MinPts; ε determines the neighborhood radius of a certain crack point in the Euclidean space, and MinPts is the minimum number of points that need to be included in the neighborhood, which is often used to judge whether a point is a core point. Based on the Euclidean distance and the neighborhood distance ε, calculate the neighborhood of each crack point. If the number of points included in the neighborhood of a crack point is greater than or equal to MinPts, then this crack point is a core point; otherwise, it is a boundary point or a noise point;

[0182] S33: Cluster expansion and core point processing: Select an unlabeled core point as the initial point, label it as the starting point of a new cluster, and assign a cluster number to this point. Subsequently, expand the neighborhood where the core point is located, and classify all points (including core points and density-reachable boundary points) in the neighborhood into the same cluster; for the newly added points in the expanded cluster, check whether they are core points; if they are core points, continue to expand the cluster; if they are boundary points, keep their cluster labels unchanged; continue to select new points from unlabeled core points for cluster expansion until all core points are processed;

[0183] That is to say, as Figure 5 shown, for core points, it is necessary to simultaneously judge whether each core point has a label and whether all core points have been read. If a core point has no label, it is necessary to expand the neighborhood and set the label, otherwise read the next core point. If it is judged that there are unread points among all core points, continue to judge whether this core point has a label, otherwise, generate all crack clusters; among them, after the core point expands the neighborhood and sets the label, it is also necessary to judge whether the newly added points in the neighborhood are core points. If they are core points, continue to expand and set the same label as the initial core point, and at the same time continue to judge whether the newly added points in the neighborhood are core points; otherwise, keep the label unchanged and generate a new crack cluster to form all crack clusters.

[0184] S34: Cluster numbering and clustering result check: According to the result of DBSCAN clustering in step S33, assign a unique crack surface number to each cluster, and each cluster represents an independent crack surface; if a certain crack point belongs to the neighborhoods of multiple core points at the same time (that is, this point is a density-reachable point), then this point will be assigned the crack surface number of the first discovered cluster; then check the clustering result to ensure that each cluster is assigned a unique crack surface number, and the number of noise points is -1; if it is found that the cluster number assignment is incorrect or missing, make necessary adjustments to ensure the accuracy of the final crack surface number;

[0185] S35: Optimization of clustering results based on magnitude: After obtaining the preliminary clustering results, the clustering results are optimized by combining the magnitude of each crack point. By visualizing the magnitude distribution of crack points within the crack surface, crack points with larger magnitudes are identified, which belong to the main crack surface; points with smaller magnitudes indicate secondary crack surfaces or crack surfaces with less expansion potential. According to the distribution characteristics of magnitudes, the division of crack points within each cluster is re-evaluated. Crack points with larger magnitudes are preferentially marked as part of the main crack surface, and crack points with smaller magnitudes are divided into secondary crack surfaces. For some crack points with larger magnitudes, if they are located in different clusters, cluster merging or re-division can be considered to better reflect the expansion potential of the cracks. The optimized clustering results will be visualized in three-dimensional space, with the main crack surface and secondary crack surfaces distinguished by different colors or sizes to assist in analyzing crack expansion characteristics.

[0186] In one example of the present invention, in the step S40, as Figure 6 shown, the B-spline surface fitting algorithm is used to reconstruct the crack points of the main crack surface to generate a smooth and continuous macroscopic crack surface, including the following steps:

[0187] S41: From the results of the clustering analysis, all valid crack points that have been classified into the same crack surface are read; among them, each crack point includes its crack spatial coordinates, crack volume, crack energy, crack spatial orientation, and crack magnitude.

[0188] S42: Since the released energy of a crack point reflects its fracture strength, and the higher the released energy, the more likely it belongs to a region with a faster expansion speed or greater expansion potential. Therefore, energy as a characteristic weight can more effectively guide the crack surface reconstruction process. At the same time, the normal direction of the crack point provides the spatial orientation information of the crack surface. Therefore, by combining energy and the normal direction, the spatial characteristics and fracture effects of the crack point can be comprehensively described.

[0189] Assume that the normal direction vector of a certain crack point is n = (n x , n y , n z ). The magnitude of this vector is calculated by the following formula:

[0190]

[0191] Since the magnitude of the normal vector itself may be affected by factors such as the position and scale of the crack point, the normal vector is normalized to ensure that the normalized normal vector only represents the direction. The calculation formula for its normalization process is as follows:

[0192]

[0193] For each crack point, calculate the comprehensive feature weight according to its fracture energy and normal direction. Assume the feature weight \(W\) of crack point \(i\) i is a combination of fracture energy and normal direction. The following weighted formula is used to calculate the comprehensive feature weight of each crack point:

[0194]

[0195] In the formula, is the modulus of the normalized normal direction, usually 1; \(E\) i is the fracture energy of the \(i\)-th crack point; \(\alpha\) and \(\beta\) are weight coefficients used to adjust the influence of the normal direction and fracture energy on the final feature weight;

[0196] S43: Based on the feature weights and three-dimensional spatial position information of the crack points in the same cluster, extract the crack points with higher feature weights through principal component analysis (PCA) as the key control points for crack surface reconstruction, so as to clarify the main direction of the crack surface. Specifically, select the two crack points with the largest feature weights as the initial control points of PCA, and define and span the main direction of the final crack surface through these points to provide the basis vectors for subsequent fitting and morphology construction;

[0197] S44: After determining the main direction of the crack surface, use the B-spline curve fitting method to fit the boundaries of the crack points in the same cluster to form a closed-loop curve as the outer contour line of the final crack surface;

[0198] S45: Triangulate the outer contour line of the already formed crack surface and divide the internal part of the final crack surface into triangular meshes;

[0199] S46: Read all the crack points in the same cluster, and combine the established triangular meshes to finally fit the crack surface to be generated to form the final macroscopic crack surface of the crack points in this cluster.

[0200] A crack surface intelligent reconstruction device based on a dynamic graph neural network according to the second aspect of the present invention includes:

[0201] A parameter acquisition module, configured to collect the original acoustic wave data through a microseismic monitoring system, perform positioning calculation and focal mechanism inversion on the original acoustic wave data, and obtain the core parameters of the coal and rock fracture source directly related to the fracture;

[0202] The model construction module is configured to preprocess the core parameters of the fracture source using a standardization processing method, construct a fracture spatio-temporal topology analysis model based on the DGCNN graph neural network, take fractures as graph nodes, and the node features include time, spatial coordinates, volume, energy, and spatial orientation; use the edge convolution mechanism to construct an adjacency graph through the spatio-temporal relationship between nodes, and aggregate neighborhood features to update the node representation, capturing local structure and global topology; optimize the parameters by minimizing the loss function, and the model outputs a high-dimensional feature vector, comprehensively reflecting spatio-temporal correlation, physical connectivity, and topological structure;

[0203] The crack surface clustering module is configured to calculate the similarity of fracture features based on the Euclidean distance, use the DBSCAN algorithm to cluster according to the density threshold, group fracture points with similar features into the same cluster, each cluster represents a crack surface, and assign a unique number, and noise points are excluded to reduce interference; and optimize the clustering results in combination with the fracture magnitude to identify the main crack surface fractures with high expansion potential;

[0204] The crack surface reconstruction module is configured to reconstruct the main crack surface using the B-spline surface fitting algorithm to generate a smooth and continuous macroscopic crack surface; select control points through the principal component analysis method to determine the main crack direction, construct a B-spline curve based on the distribution of fracture points to generate a boundary contour, triangulate the outer contour line and divide the grid, combine the fracture points and the triangular grid, and use the B-spline algorithm to accurately fit the fracture surface.

[0205] This system obtains the core parameters of the focal mechanism through microseismic monitoring inversion, which can directly reflect the source characteristics of fracture points, and has high reliability and accuracy. Compared with the traditional method that only relies on the characterization of spatial positioning points, the core parameters provided by the focal mechanism inversion as data input can more accurately capture the spatio-temporal correlation and physical connectivity between fractures, significantly improving the scientificity and reliability of the intelligent reconstruction of fracture surfaces. At the same time, it is closer to the true evolution law of internal fractures in rock masses, laying a solid data foundation for analyzing the fracture propagation and penetration mechanism.

[0206] This system constructs a fracture spatio-temporal topology analysis model based on the dynamic graph neural network, captures the spatio-temporal correlation and global topology characteristics of fracture points through edge convolution, and uses the adjacency relationship to deeply explore the spatio-temporal correlation, physical connectivity, and topological structure between fracture points, breaking through the limitations of traditional methods in dealing with complex fracture networks, significantly improving the accuracy of fracture network analysis and the topological characterization ability, and providing strong technical support for the intelligent reconstruction of fracture surfaces.

[0207] To overcome the deficiencies of traditional fracture characterization methods in analyzing spatio-temporal topological correlations and penetration mechanisms, the system improves the adjacency relationship in the graph neural network. On the basis of considering the relative relationships of the spatial positions and time series of fracture points, it further integrates the spatial orientation and energy dissipation characteristics of fracture points, deeply explores the physical connectivity and topological structure between fracture points, and enables it to more effectively capture and analyze the complex spatio-temporal topological relationships between fracture points.

[0208] By combining graph neural network modeling, DBSCAN clustering, and B-spline fitting, the system can accurately reconstruct the geometric morphology and dynamic evolution characteristics of the fracture surface, intuitively display the expansion process, temporal evolution, and spatial distribution of the macroscopic fracture surface, greatly improving the accuracy and reliability of fracture surface inversion, providing advanced technical support for in-depth analysis of the rock mass fracture evolution mechanism, and at the same time providing a more scientific basis for the stability assessment and disaster warning of rock mass engineering.

[0209] In the foregoing, the exemplary embodiments of the fracture surface reconstruction method and device based on the dynamic graph neural network proposed by the present invention have been described in detail with reference to the preferred embodiments. However, those skilled in the art can understand that, without departing from the concept of the present invention, various modifications and variations can be made to the above specific embodiments, and various combinations can be made to the various technical features and structures proposed by the present invention, without exceeding the protection scope of the present invention. The protection scope of the present invention is determined by the appended claims.

Claims

1. A fracture surface reconstruction method based on dynamic graph neural network, characterized in that: The steps include: S10: acquiring original acoustic wave data through the microseismic monitoring system, performing positioning calculation and focal mechanism inversion on the original acoustic wave data, and obtaining the core parameters of the coal-rock fracture source directly related to the fracture; wherein the core parameters of the coal-rock fracture source include: fracture space coordinates, fracture volume, fracture energy, fracture space orientation and fracture magnitude; S20: A standardized processing method is used to preprocess the core parameters of the rupture source, and a fracture spatiotemporal topological analysis model is constructed based on the DGCNN graph neural network, with fractures as graph nodes. The edge convolution mechanism is used to construct an adjacency graph through the spatiotemporal relationship between nodes, and the neighborhood features are aggregated to update the node representation to capture the local structure and global topology. S30: Calculate the similarity of crack characteristics based on Euclidean distance, use DBSCAN algorithm to cluster by density threshold, classify the crack points with similar characteristics into the same cluster, each cluster represents a crack surface and is assigned a unique number; and optimize the clustering results in combination with the crack magnitude to identify the main crack surface cracks with high expansion potential; S40: The main crack surface is reconstructed using the B-spline surface fitting algorithm to generate a smooth and continuous macro crack surface. The main crack direction is determined by selecting control points using the principal component analysis method. The boundary contour is generated by constructing a B-spline curve based on the distribution of crack points. The outer contour is triangulated and meshed. The crack points and triangular mesh are combined to accurately fit the crack surface using the B-spline algorithm.

2. The fracture surface reconstruction method based on dynamic graph neural network according to claim 1 is characterized in that: In the step S10, the original acoustic wave data is subjected to positioning calculation and focal mechanism inversion to obtain the focal mechanism core parameters directly related to the rupture, including the following steps: S11: Post-process the original acoustic wave data collected by the microseismic monitoring system and extract the first wave arrival time and amplitude information using the red-chi information criterion method; S12: Using the arrival time data of the first wave of the acoustic wave at sensors at different positions of the microseismic monitoring system, the simplex positioning robust algorithm is used to calculate the spatial positioning of the acoustic emission source to obtain the spatial coordinates of the fracture; S13: Based on the constraints of the tensile-shear rupture source model, the six components M of the source rupture moment tensor are solved by the acoustic first wave amplitude data collected by sensors at different positions. pq ; S14: According to the calculated moment tensor components, quantitative inversion is performed to calculate the core parameters of the coal rock fracture source, wherein the core parameters of the coal rock fracture source include fracture volume, fracture energy, fracture spatial orientation and fracture magnitude.

3. The fracture surface reconstruction method based on dynamic graph neural network according to claim 2 is characterized in that: In step S14, the specific calculation formula of the core parameters of the coal rock fracture source is as follows: Fracture volume: Where ΔV is the fracture volume of the crack point, M1 and M3 are the moment tensors M pq The characteristic value of , μ is the Lamé constant; Fracture Space Orientation: n=(cosα,0,±sinα) Where α is the angle between the movement direction of the earthquake source rupture surface and the normal direction, M1, M2, and M3 are the moment tensors M pq The eigenvalues ​​of , μ and λ are the Lame constants, n is the spatial orientation, and b is the direction of motion; Rift Energy: Where: ΔA is the fracture area of ​​the crack point, σ is the material strength, where σ t is the tensile strength, σ s is the shear strength; Fissure Magnitude: Where M1, M2, and M3 are moment tensors M pq The characteristic value of .

4. The fracture surface reconstruction method based on dynamic graph neural network according to claim 1 is characterized in that: In step S20, the core parameters of the rupture source are preprocessed using a standardized processing method, which specifically includes the following: Assume that the preprocessed rupture source core parameters are x1, x2, …, x n-1 , x n , mapped to the interval [0, 1] by the minimum-maximum normalization method, and its normalization formula is as follows: In the formula, is the minimum value of the preprocessing parameter, is the maximum value of the preprocessing parameter, y i ∈[0,1].

5. The fracture surface reconstruction method based on dynamic graph neural network according to claim 1 is characterized in that: In step S20, the crack spatiotemporal topology analysis model includes a spatial transformation module, an edge convolution module, a maximum pooling layer and a fully connected layer. The spatial transformation module is configured to normalize and align the input point cloud data, and obtain an affine transformation matrix after learning, which is used to transform the input point cloud data into a standard coordinate system; The edge convolution module is configured to perform convolution operations on point cloud data for feature extraction and modeling local structural information, dynamically learning and extracting local structural features; The maximum pooling layer is configured to aggregate the features of the crack points and extract the most representative local information; The fully connected layers are configured to map global features to a higher dimensional space in order to capture complex nonlinear patterns and potential relationships in the data.

6. The method for reconstructing crack surfaces based on dynamic graph neural network according to claim 5, characterized in that: Perform convolution operations on point cloud data to extract features and model local structural information, dynamically learn and extract local structural features, specifically including the following steps: (a) Point cloud data input: Assuming that the input point cloud data contains n crack points, the point cloud data X can be expressed as: X={x1,x2,....x n }∈R D Where: x i ={px i ,py i ,pz i ,t i ,nx i ,ny i ,nz i ,V i ,E i } is the characteristic of the i-th crack point; D is the feature dimension; px i ,py i ,pz i } is the spatial coordinate of the i-th crack point; t i is the timestamp of the i-th crack point; {nx i ,ny i ,nz i } is the spatial orientation of the i-th crack point; V i is the rupture volume of the i-th crack point; E i is the released energy of the i-th crack point; (b) Establishing adjacency relationships: The adjacency relationship is used to establish a graph structure between point clouds. Each fracture point is regarded as a node in the graph, and the adjacency relationship between nodes is dynamically constructed according to predefined rules. Among them, whether two fracture points are connected is determined based on temporal adjacency, spatial topological relationship, spatial orientation similarity, and energy similarity. (c) Build graph structure: Define a directed graph G for representing the local point cloud structure. G consists of vertices V and directed edges E, and its expression is: G=(V,E),E∈V×V Where: V i ={x1,x2,....x n } is the set of all positioning points, Represents x i The set of directed edges to its k adjacent vertices; (d) Edge feature extraction: The local features are obtained by calculating the difference between the positioning point and its adjacent points, and a multi-layer perceptron is used. Θ Extract edge features, assuming that the positioning point x i Its jth neighbor The edge features between are: Where: f Θ is a set of nonlinear functions parameterized by a set of learnable parameters Θ; (e) Feature aggregation: After the edge features of each center point are extracted, they are first linearly transformed through a fully connected layer and then normalized. Then, the features of the center point are aggregated with the features of the points in its local neighborhood through an aggregation function to obtain the aggregated features. The input center point features are added to the aggregated center point features to achieve residual connection. Finally, the ReLU activation function is used to achieve nonlinear mapping to obtain the output of the edge convolutional network.

7. The fracture surface reconstruction method based on dynamic graph neural network according to claim 1 is characterized in that: In step (b) of establishing an adjacency relationship, determining whether two fracture points are connected requires that the following conditions are met at the same time: Temporal proximity: The time difference between two positioning points should be less than a time threshold, indicating that they occurred close in time and may belong to the same crack growth process: t i -t j |≤Δt Where: t i ,t j are the rupture times of crack points i and j, respectively, and Δt is the predefined time threshold; Spatial topological relationship: By calculating the geometric relationship between crack points, three topological relationships are defined: intersection, adjacency, and separation. Only when two crack points intersect or are adjacent can an adjacency relationship be established, indicating that they are close enough in space and can be connected, and belong to the same crack; Among them, the spatial distance d between the crack point i and other crack points j is calculated ij : Where: (x i ,y i ,z i )、(x j ,y j ,z j ) are the positioning space coordinates of crack points i and j respectively; Spatial orientation similarity: The spatial orientation vectors of two positioning points need to be similar within a certain error range, then they are considered to belong to the same crack. The similarity of the directions is determined by calculating the angle between the direction vectors, where the angle between the direction vectors is expressed as: Where n i =(nx i ,ny i ,nz i ) and n j =(nx j ,ny j ,nz j ) are the spatial orientation vectors of the two crack points, and Δθ is the predefined angle error range; Energy similarity: The energy difference between two positioning points should be less than a certain threshold, indicating that they have similar energy release patterns and can be physically connected or interlinked. The judgment expression is: |And i -AND j |≤ΔE Where: E i 、E j are the release energy values ​​of crack points i and j respectively, and ΔE is the predefined exponential threshold.

8. The fracture surface reconstruction method based on dynamic graph neural network according to claim 1 is characterized in that: In step S30, clustering is performed according to the density threshold using the DBSCAN algorithm, including the following steps: S31: Feature vector reading and Euclidean distance calculation: Read the high-dimensional feature vector of each crack point obtained by the crack spatiotemporal topological analysis model, and calculate the Euclidean distance between the crack points based on the high-dimensional feature vector to measure the similarity between the crack points; for each pair of crack points P i =(x1,x2,x3....x n ) and P j =(y1,y2,y3....y n ), the Euclidean distance calculation formula is as follows: Where n is the dimension of the feature vector, x k and k are the values ​​of the kth dimension in the high-dimensional feature vectors of the two crack points respectively; S32: Determine the neighborhood distance and the minimum number of points: Select an appropriate neighborhood distance ε and a minimum number of points MinPts; Calculate the neighborhood of each crack point based on the Euclidean distance and the neighborhood distance ∈. If the number of points contained in the neighborhood of a crack point is greater than or equal to MinPts, then the crack point is a core point; otherwise, it is a boundary point or a noise point; S33: Cluster expansion and core point processing: Select one of the unmarked core points as the initial point, mark it as the starting point of a new cluster, and assign a cluster number to the point; expand the neighborhood of the core point, and classify all points in the neighborhood into the same cluster; for the newly added point in the expanded cluster, check whether it is a core point; if it is a core point, continue to expand the cluster; if it is a boundary point, keep its cluster label unchanged; continue to select new points from the unmarked core points for cluster expansion until all core points are processed; S34: Check cluster number and clustering result: According to the result of DBSCAN clustering in step S33, a unique crack surface number is assigned to each cluster, and each cluster represents an independent crack surface; if a crack point belongs to the neighborhood of multiple core points at the same time, the point will be assigned to the crack surface number of the first cluster found; then check the clustering results to ensure that each cluster is assigned a unique crack surface number, where the number of noise points is -1; if it is found that the cluster number assignment is incorrect or omitted, make necessary adjustments to ensure that the final crack surface number is accurate; S35: Optimization of clustering results based on magnitude: After obtaining the preliminary clustering results, the clustering results are optimized in combination with the magnitude of each crack point. By visualizing the magnitude distribution of the crack points within the crack surface, the crack points with larger magnitudes are identified, which belong to the main crack surface; points with smaller magnitudes represent secondary crack surfaces or crack surfaces with smaller expansion potential.

9. The method for reconstructing crack surfaces based on dynamic graph neural network according to claim 1, characterized in that: In step S40, the crack points of the main crack surface are reconstructed using a B-spline surface fitting algorithm to generate a smooth and continuous macro crack surface, including the following steps: S41: reading all valid crack points that have been classified as the same crack surface from the results of cluster analysis; wherein each crack point includes crack space coordinates, crack volume, crack space orientation, crack energy and crack magnitude; S42: Combine energy and normal direction to fully describe the spatial characteristics of the crack point and the rupture impact; Assume that the normal direction vector of a crack point is n=(n x ,n y ,n z ), the formula for calculating the modulus of the vector is as follows: Normalize the normal vector to ensure that the normalized normal vector only represents the direction. The calculation formula of the normalization process is as follows: For each crack point, the comprehensive feature weight is calculated according to its fracture energy and normal direction; assuming that the feature weight W of crack point i is i is a combination of fracture energy and normal direction, then the calculation formula for the comprehensive characteristic weight of each crack point is as follows: In the formula, is the normalized modulus in the normal direction; E i is the fracture energy of the i-th crack point; α and β are weight coefficients; S43: Based on the feature weights and three-dimensional spatial position information of the same-cluster fracture points, the fracture points with higher feature weights are extracted through principal component analysis as key control points for fracture surface reconstruction, thereby clarifying the main direction of the fracture surface; S44: after determining the main direction of the crack surface, the boundaries of the same cluster of crack points are fitted using a B-spline curve fitting method to form a closed loop curve as the outer contour line of the final crack surface; S45: triangulating the outer contour of the formed crack surface, and dividing the interior of the final crack surface into triangular meshes; S46: Read all the crack points in the same cluster, and perform final fitting on the crack surface to be generated in combination with the established triangular mesh to form a final macro crack surface of the crack points in the same cluster.

10. A fracture surface reconstruction device based on dynamic graph neural network, characterized in that: include: A parameter acquisition module is configured to acquire original acoustic wave data through a microseismic monitoring system, perform positioning calculation and focal mechanism inversion on the original acoustic wave data, and acquire core parameters of the coal and rock fracture source directly related to the fracture; The model building module is configured to pre-process the core parameters of the rupture source using a standardized processing method, build a fracture spatiotemporal topology analysis model based on the DGCNN graph neural network, and use fractures as graph nodes; use the edge convolution mechanism to build an adjacency graph through the spatiotemporal relationship between nodes, and aggregate neighborhood features to update the node representation to capture local structure and global topology; The crack surface clustering module is configured to calculate the similarity of crack features based on Euclidean distance, cluster the crack points with similar features into the same cluster by density threshold using DBSCAN algorithm, each cluster represents a crack surface and is assigned a unique number; and optimize the clustering results in combination with the crack magnitude to identify the main crack surface cracks with high expansion potential; The crack surface reconstruction module is configured to reconstruct the main crack surface using the B-spline surface fitting algorithm to generate a smooth and continuous macro crack surface; the control points are selected through the principal component analysis method to determine the main direction of the crack, and the B-spline curve is constructed based on the distribution of crack points to generate the boundary contour. The outer contour line is triangulated and meshed, and the crack points and triangular meshes are combined to use the B-spline algorithm to accurately fit the crack surface.

Images