Engineering rock mass fissure surface dynamic extension prediction method and system

Abstract

The invention discloses an engineering rock mass fracture surface dynamic extension prediction method and system, and the method comprises the following steps: obtaining original sound wave data, carrying out the positioning calculation and seismic source mechanism inversion of the original sound wave data, obtaining the core parameters of a coal rock fracture source at each moment, and calculating a fracture penetration possibility index BCI; fracture similarity is calculated through multi-dimensional features, density clustering is performed by using a DBSCAN clustering algorithm, and a reconstructed fracture surface time sequence image is obtained in combination with BCI optimization and a B spline surface fitting technology. A fracture surface image spatio-temporal data set is obtained through preprocessing and is divided, an improved SimVP spatio-temporal neural network is utilized to construct a fracture surface image sequence prediction model for training, the multi-head attention mechanism and local residual error connection are introduced into an encoder and a decoder, a Poolform network is used as a translator, and a fracture surface image sequence prediction model is constructed through the improved SimVP spatio-temporal neural network. And finally, utilizing the trained prediction model to accurately predict the dynamic expansion of the fracture surface at the future moment.

Description

Technical Field

[0001] The present invention relates to the technical field of fracture prediction, and in particular to a method and system for dynamically predicting the expansion of engineering rock mass fracture surfaces. Background Art

[0002] As an important engineering measure, the hydrofracturing measure has the advantages of pressure relief, permeability enhancement, and property modification, and has been widely applied to engineering such as deep oil and gas resource exploitation, hot dry rock development and utilization, and carbon storage layer transformation. During the implementation of the hydrofracturing measure, the expansion characteristics and spatial distribution of fractures are related to the implementation effect of the measure and engineering safety. By accurately predicting the dynamic expansion process of fractures, the exploitable property of the reservoir can be effectively predicted, the design of hydraulic fracturing measures can be optimized, and the risk of induced seismicity during fracturing can be reduced, thereby improving the economic efficiency and safety of green resource development.

[0003] Traditional fracture expansion prediction methods mostly rely on physical models and empirical formulas. These methods often assume that fracture expansion follows certain fixed laws and are difficult to fully capture the complexity of the underground environment and the dynamic changes of fracture expansion. Therefore, as an advanced data-driven technology, deep learning has gradually become a research hotspot in the field of fracture expansion prediction. Patent CN119228748A provides a method, device, and storage medium for predicting rock mass fractures. By constructing a data set and using an image interpolation algorithm to optimize data continuity, a spatio-temporal prediction network uses a self-attention mechanism to capture the spatial and temporal dependence relationships between feature maps, achieving accurate capture of the dynamic changes of fractures. Patent CN118506104A discloses a method and system for rock fracture identification and expansion prediction. The processed CT scan slice images are input into the SAM segmentation model for rock fracture identification, and a graph convolutional neural network is established to analyze the connection relationships between fractures and predict the fracture expansion path. Patent CN117557874A proposes an intelligent prediction method for the expansion of coal-rock fatigue deformation fractures. The CT images of coal-rock specimens are used as a time-series data set, and a regression model for the expansion of coal-rock fatigue fractures is constructed to predict the development and expansion of pore fractures and the deformation of coal-rock in real time. Although the above methods have made significant progress in fracture expansion prediction, they still face the challenge of processing high-dimensional and complex spatio-temporal data. In this context, the SimVP spatio-temporal neural network, as a video prediction network completely based on the CNN architecture, with its simple and efficient design, can better handle the complexity of spatio-temporal data and provides a new technical path for the dynamic prediction of fracture expansion. Summary of the Invention

[0004] In response to the problems and requirements raised above, this solution proposes a method and system for dynamically predicting the expansion of engineering rock mass fracture surfaces. Due to the following technical features, the above technical objectives can be achieved, and many other technical effects can be brought.

[0005] An object of the present invention is to propose a method for predicting the dynamic propagation of fracture surfaces in engineering rock masses, which is characterized by including the following steps:

[0006] S10: Obtain the original acoustic wave data, perform positioning calculations and focal mechanism inversions on it, obtain the core parameters of the coal-rock fracture source directly related to fracture at each moment, and calculate the fracture penetration possibility index BCI of the fracture at each moment in combination with the standardized fracture volume, source radius, and spatial distance;

[0007] S20: Calculate the similarity between fractures at each moment according to the multi-dimensional characteristics of the fractures, use the DBSCAN clustering algorithm to perform density clustering on the fractures, optimize and identify potential fracture surfaces with high connectivity and expansion potential in combination with the fracture penetration possibility index BCI, and use the B-spline surface fitting algorithm to smoothly fit the fracture surface to obtain the reconstructed fracture surface image;

[0008] S30: Arrange the reconstructed fracture surface images in chronological order to form a fracture surface time series image; after preprocessing, construct a spatio-temporal data set of fracture surface images, and divide it into a training set, a validation set, and a test set;

[0009] S40: Use an improved SimVP spatio-temporal neural network to construct a fracture surface image sequence prediction model;

[0010] S50: Input the training set into the fracture surface image sequence prediction model for training, use the validation set to test and evaluate the trained prediction model, and continuously adjust the parameters for training; after training is completed, input the fracture surface sequence images of the test set into the trained prediction network to obtain the predicted fracture surface image sequence, and use the corresponding evaluation indicators to systematically evaluate the prediction results.

[0011] In addition, according to the method for predicting the dynamic propagation of fracture surfaces in engineering rock masses of the present invention, it may also have the following technical features:

[0012] In an example of the present invention, in the step S10, perform positioning calculations and focal mechanism inversions on the original acoustic wave data, obtain the core parameters of the focal mechanism directly related to fracture at each moment, and calculate the fracture penetration possibility index BCI of the fracture at each moment, including the following steps:

[0013] 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;

[0014] S12: Use the first arrival time data 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;

[0015] S13: Solve the six components \(M\) of the source rupture moment tensor based on the acoustic wave first arrival amplitude data collected by sensors at different positions under the constraint conditions of the strike-slip rupture source model. pq ;

[0016] S14: Quantitatively invert and calculate the core parameters of the coal and rock fracture source according to the calculated moment tensor components. Among them, the core parameters of the coal and rock fracture source include the fracture spatial orientation, fracture volume, and fracture energy.

[0017] S15: Determine the critical distance \(D\) for fracture penetration according to the source radius of the fracture point. i , and convert the data into a standard normal distribution with zero mean and unit variance through the Z - score standardization method.

[0018] S16: When two adjacent fracture points \(i\) and \(j\) meet the fracture penetration condition, calculate the fracture penetration possibility index \(BCI\) between the two adjacent fracture points. ij , and calculate the total fracture penetration index \(BCI\) of the fracture point \(i\) therefrom. i .

[0019] In an example of the present invention, in the step S20, the DBSCAN algorithm is used to perform density clustering on the fractures, including the following steps:

[0020] S211: Multidimensional feature vector reading and Euclidean distance calculation: Read the multidimensional feature vectors of each fracture point obtained by source mechanism inversion at each moment, and calculate the Euclidean distance between the fracture points.

[0021] S212: Determine the neighborhood distance and the minimum number of points: Select an appropriate neighborhood distance \(\epsilon\) and the minimum number of points \(MinPts\). Based on the Euclidean distance and the neighborhood distance \(\epsilon\), calculate the neighborhood of each fracture point; if the number of points contained in the neighborhood of a fracture point is greater than or equal to \(MinPts\), then this fracture point is a core point; otherwise, it is a boundary point or a noise point.

[0022] S213: 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; then, expand the neighborhood where the core point is located, and classify all points in the neighborhood into 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 boundary 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.

[0023] S214: Cluster number and clustering result check: According to the result of DBSCAN clustering, assign a unique crack surface number to each cluster, where each cluster represents an independent crack surface; if a fissure point belongs to the neighborhoods of multiple core points simultaneously, 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.

[0024] In an example of the present invention, in the step S20, optimize and identify potential fissure surfaces with high penetrability and expansion potential by combining the breakage penetration possibility index BCI, and apply the B-spline surface fitting algorithm to smoothly fit the fissure surfaces to obtain a reconstructed fissure surface image, including the following steps:

[0025] S221: Optimize and screen fissures based on BCI values: For each fissure cluster, calculate the breakage penetration possibility index BCI of each fissure therein, and further visualize the BCI value distribution of the fissures, set an appropriate threshold, and select fissure points with higher BCI values as potential fissure surfaces;

[0026] S222: Reconstruct the time series image of the fissure surface: Read the fissure data obtained by clustering analysis at all times, combine the characteristic information of the spatial coordinates, BCI values, volumes, and normal directions of the fissures at each time, and perform fitting processing on the fissures at each time; through the B-spline curve and triangular mesh division method, accurately reconstruct the fissure surface at each time, and finally integrate the reconstruction results of the fissure surfaces at all times to generate a complete time series image of the fissure surface.

[0027] In an example of the present invention, in the step S30, preprocess the data and divide the data set, including:

[0028] Convert the breakage penetration possibility index of each fissure point on the fissure surface obtained by clustering fitting into the corresponding gray value, and interpolate it onto a uniform 128×128×128 voxel grid, perform normalization processing on the interpolated image, and construct a spatio-temporal data set of the fissure surface image. For example, use 80% of the data set as the training set, 10% as the validation set, and 10% as the test set.

[0029] In an example of the present invention, in the step S40, the fissure surface image sequence prediction model includes a spatial encoder, a translator, and a spatial decoder,

[0030] The spatial encoder is configured to map the input high-dimensional fissure map data to a low-dimensional latent space, use three-dimensional convolution operations to capture the spatial features in the three-dimensional fissure map data, and introduce a multi-head attention mechanism before convolution to capture spatio-temporal dependencies;

[0031] The translator is configured to use the Poolformer network as the prediction network translator to learn the spatial dependencies and temporal variations of the crack surface image sequence from the latent space. The output tensor of the encoder is used as the input of the Poolformer. Then, multiple sequence frames are stacked along the time axis direction, and the Poolformer is used to learn from the stacked multi-frame features to capture the inherent temporal evolution law within the crack surface image sequence data. Finally, the output result is used as the input of the decoder;

[0032] The spatial decoder is configured to decode the information in the latent space into the predicted future crack surface image, use three-dimensional deconvolution operations to gradually restore the details of the image, and introduce a multi-head attention mechanism before deconvolution to capture the spatio-temporal dependencies of the decoder features.

[0033] In an example of the present invention, the spatial encoder includes:

[0034] The first multi-head attention mechanism module is configured to map the input original crack map features into different feature space queries Q, keys K, and values V, and pack them into matrices Q 1 , K 1 , V 1 . Q 1 , K 1 , V 1 are linearly transformed with different weight matrices to obtain Q 1 ’, K 1 ’, V 1 ’. Then, for each attention head, its attention value is calculated independently; finally, the outputs of all heads are concatenated and passed through a final linear transformation to obtain the final multi-head attention output;

[0035] The first local residual connection module is provided at the input and output ends of the first multi-head attention mechanism module, and is configured to maintain the transmission of spatial information;

[0036] The three-dimensional convolution module includes multiple cascaded convolution layers. The convolution layer includes: convolution, normalization operation, and non-linear transformation. The three-dimensional convolution module is configured to perform feature extraction on z′ 0 output by the first multi-head attention mechanism module after passing through the first local residual connection module, where a downsampling operation is performed every two convolution layers.

[0037] In an example of the present invention, the spatial decoder includes:

[0038] The second multi-head attention mechanism module is configured to use the output tensor w of the translator 0Map to different feature spaces for queries Q, keys K, and values V, and pack them into matrices Q 2 , K 2 , V 2 . Then, Q 2 , K 2 , V 2 are linearly transformed with different weight matrices to obtain Q 2 ', K 2 ', V 2 '; Next, for each attention head, calculate its attention value independently; finally, concatenate the outputs of all heads and perform a final linear transformation to obtain the final multi-head attention output;

[0039] Second local residual connection module. Set second local residual connection modules at the input and output ends of the second multi-head attention mechanism module, configured to maintain the transmission of spatial information;

[0040] 3D transposed convolution module, including multiple cascaded transposed convolution layers. The transposed convolution layer includes: transposed convolution, normalization operation, and non-linear transformation. The 3D transposed convolution module is configured to perform row feature extraction and spatial resolution restoration on the output w' of the second multi-head attention mechanism module passing through the second local residual connection module 0 , where an upsampling operation is performed using PixelShuffle for every two transposed convolution layers.

[0041] In an example of the present invention, step S50 specifically includes the following steps:

[0042] S51: Train the crack surface image sequence prediction model with a training set, adjust the batchsize, learning rate, optimizer, and number of epochs, and use the L2 loss function and the motion perception loss function L m to construct the total loss function L to supervise the training of the model;

[0043] S52: Input the test set into the trained crack surface image sequence prediction model to obtain the predicted crack surface image sequence, and evaluate the prediction results using evaluation metrics such as MSE, SSIM, and PSNR. The evaluation metrics are as follows:

[0044]

[0045] In the formula, Y i and respectively represent the true value and the predicted value at the i-th time step, μ Y and respectively represent the averages of Y i and , σ Y and respectively represent Yi and standard deviation of, denotes Y i and covariance of, c 1 and c 2 are constants, I MAX denotes the maximum pixel value of the image.

[0046] In an example of the present invention, in the step S52, the test set is input into the trained fracture surface image sequence prediction model to obtain the predicted fracture surface image sequence, which specifically includes the following steps:

[0047] Input the fracture surface image sequence X ∈ R T×C×H×W×D into the prediction model, where T represents the time step, C represents the number of channels of each image, and H, W, D respectively represent the height, width and depth of each image.

[0048] A set of dynamic sequences is formed according to the change X of T time steps. The fracture surface image sequence prediction predicts the future T' frames based on the given T-frame images, and its modeling formula is as follows:

[0049]

[0050] In the formula: X T×C×H×W×D is the input fracture surface image sequence, is the output predicted fracture surface image sequence, Y T′×C×H×W×D is the real fracture surface image sequence, and Θ represents a series of parameters to be optimized;

[0051] Another object of the present invention is to propose an engineering rock mass fracture surface dynamic expansion prediction system, including:

[0052] A parameter acquisition and calculation module, configured to obtain the original acoustic wave data, perform positioning calculation and seismic source mechanism inversion on it, obtain the core parameters of the coal and rock fracture source directly related to the fracture at each moment, and calculate the fracture penetration possibility index BCI of the fracture at each moment in combination with the standardized fracture volume, seismic source radius and spatial distance;

[0053] A fracture surface image reconstruction module, configured to calculate the similarity between fractures at each moment according to the multi-dimensional characteristics of the fractures, perform density clustering on the fractures using the DBSCAN clustering algorithm, optimize and identify the potential fracture surfaces with higher penetration and expansion potential in combination with the fracture penetration possibility index BCI, and apply the B-spline surface fitting algorithm to smooth and fit the fracture surfaces to obtain the reconstructed fracture surface image;

[0054] The data set processing and partitioning module is configured to arrange the reconstructed fracture surface images in chronological order to form fracture surface time series images; after preprocessing, construct a fracture surface image spatiotemporal data set and divide it into a training set, a validation set and a test set;

[0055] A prediction model building module, configured to build a crack surface image sequence prediction model using an improved SimVP spatiotemporal neural network;

[0056] The fracture surface image prediction module is configured to input the training set into the fracture surface image sequence prediction model for training, use the validation set to test and evaluate the trained prediction model, and continuously adjust the parameter training; after the training is completed, the fracture surface sequence image of the test set is input into the trained prediction network to obtain the predicted fracture surface image sequence, and the prediction results are systematically evaluated using corresponding evaluation indicators.

[0057] Compared with the prior art, this technical solution has the following beneficial effects:

[0058] The present invention obtains the core parameters of the focal mechanism through microseismic monitoring inversion, can accurately capture the source characteristics of the cracks, and use them as input data for the DBSCAN clustering algorithm, which significantly improves the accuracy and reliability of the reconstruction of the time series image of the crack surface, and provides more scientific technical support for further research on the dynamic expansion prediction of cracks.

[0059] Different from the traditional time series prediction model, the SimVP network adopted in the present invention is a video prediction network completely based on the CNN architecture. It has no complex loop structure or additional modules, avoids the introduction of too many techniques and strategies, and thus achieves efficient prediction of the crack expansion process while keeping the network structure simple.

[0060] When constructing a crack surface image sequence prediction model, the present invention introduces a multi-head attention mechanism in the encoder and decoder, which effectively extracts the high-dimensional features of the crack surface image sequence data, enabling the model to maintain the spatial information of the sequence while learning the temporal correlation, thereby improving the accuracy and reliability of crack surface prediction.

[0061] The present invention fully integrates multiple technical advantages through the comprehensive application of focal mechanism inversion, DBSCAN clustering algorithm and spatiotemporal neural network prediction model, effectively capturing the spatiotemporal correlation in the process of fracture expansion. The prediction model can deeply explore the time series characteristics and spatial distribution patterns in the process of fracture expansion, significantly improving the prediction accuracy of the dynamic evolution of the fracture surface, and providing a basis for optimizing hydraulic fracturing design and implementation.

[0062] The best embodiment for carrying out the present invention will be described in more detail below with reference to the accompanying drawings so that the features and advantages of the present invention can be easily understood. Brief Description of the Drawings

[0063] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments of the present invention will be briefly introduced below. Among them, the drawings are only used to show some embodiments of the present invention, rather than limiting all embodiments of the present invention thereto.

[0064] Figure 1 It is a flowchart of a method for predicting the dynamic expansion of a fracture surface of an engineering rock mass according to an embodiment of the present invention;

[0065] Figure 2 It is a flowchart of acoustic wave data location inversion calculation according to an embodiment of the present invention;

[0066] Figure 3 It is a DBSCAN clustering flowchart according to an embodiment of the present invention;

[0067] Figure 4 It is a schematic flowchart of a prediction model for a fracture surface image sequence according to an embodiment of the present invention;

[0068] Figure 5 It is a schematic structural diagram of a prediction model for a fracture surface image sequence according to an embodiment of the present invention;

[0069] Figure 6 It is a structural diagram of a multi-head attention mechanism module according to an embodiment of the present invention. Detailed Embodiments

[0070] 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 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.

[0071] Unless otherwise defined, the technical terms or scientific terms used herein shall have the ordinary meanings understood by those of ordinary skill in the art to which this invention pertains. The terms "first", "second" and similar terms used in the description 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, similar terms such as "a" or "an" do not necessarily denote a quantity limitation. 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 positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0072] According to an engineering rock mass fracture surface dynamic expansion prediction method of the first aspect of the present invention, as Figure 1 shown, the method includes the following steps:

[0073] S10: As Figure 2 shown, by extracting the original acoustic wave data collected by the microseismic monitoring system and performing positioning calculation and focal mechanism inversion on it, the core parameters of the coal-rock fracture source directly related to the fracture at each moment are obtained, and the fracture penetration possibility index BCI of the fracture at each moment is calculated by combining the standardized fracture volume, source radius and spatial distance;

[0074] S20: Calculate the similarity between fractures at each moment according to multi-dimensional characteristics such as the spatial distance, orientation, volume and energy of the fractures, and use the DBSCAN clustering algorithm to perform density clustering on the fractures. Combine the fracture penetration possibility index BCI to optimize and identify potential fracture surfaces with high penetration and expansion potential, and use the B-spline surface fitting algorithm to smoothly fit the fracture surfaces to obtain a reconstructed fracture surface image;

[0075] S30: Arrange the reconstructed fracture surface images in chronological order to form a fracture surface time series image; after preprocessing, construct a fracture surface image spatio-temporal data set, and divide it into a training set, a validation set and a test set;

[0076] S40: Use an improved SimVP spatio-temporal neural network to construct a fracture surface image sequence prediction model, including introducing a multi-head self-attention mechanism and local residual connections in the encoder and decoder, and replacing the translator with a Poolformer network;

[0077] S50: Input the training set into the fracture surface image sequence prediction model for training, use the validation set to test and evaluate the trained prediction model, and continuously adjust the parameters for training. After training is completed, input the fracture surface sequence images of the test set into the trained prediction network to obtain the predicted fracture surface image sequence, and use the corresponding evaluation indicators to systematically evaluate the prediction results.

[0078] This prediction method obtains the core parameters of the focal mechanism through microseismic monitoring inversion, can accurately capture the source characteristics of fractures, and uses them as input data for the DBSCAN clustering algorithm, significantly improving the accuracy and reliability of the reconstruction of fracture surface time series images, and providing more scientific technical support for further research on the prediction of fracture dynamic expansion.

[0079] Different from traditional time series prediction models, the SimVP network adopted by this prediction method is a video prediction network completely based on the CNN architecture, without complex loop structures or additional modules, avoiding the introduction of too many techniques and strategies, thus achieving efficient prediction of the fracture expansion process while keeping the network structure simple.

[0080] When constructing the fracture surface image sequence prediction model, this prediction method introduces the multi-head attention mechanism in the encoder and decoder, effectively extracts the high-dimensional features of the fracture surface image sequence data, enables the model to maintain the spatial information of the sequence while learning the time series correlation, thereby improving the accuracy and reliability of fracture surface prediction.

[0081] Through the comprehensive application of focal mechanism inversion, DBSCAN clustering algorithm and spatio-temporal neural network prediction model, this prediction method fully integrates the advantages of multiple technologies and effectively captures the spatio-temporal correlation in the fracture expansion process. And the prediction model can deeply mine the time series features and spatial distribution patterns in the fracture expansion process, significantly improving the prediction accuracy of the dynamic evolution of the fracture surface, providing a basis for optimizing the design and implementation of hydraulic fracturing.

[0082] 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 at each moment, and calculate the fracture penetration probability index BCI at each moment, including the following steps:

[0083] 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.

[0084] S12: Use the first arrival time data of acoustic waves at sensors in different positions of the microseismic monitoring system, and use the simplex positioning robust algorithm to perform spatial positioning calculation on the acoustic emission source.

[0085] S13: Solve the six components \(M\) of the source rupture moment tensor from the first arrival amplitude data of acoustic waves collected by sensors at different positions under the constraint conditions of the Zhang-Jian rupture source model pq ;

[0086] S14: According to the calculated moment tensor components, quantitatively inversely calculate the core parameters of the coal-rock fracture source, where the core parameters of the coal-rock fracture source include fracture spatial orientation, fracture volume, and fracture energy;

[0087] S15: Determine the critical distance \(D\) for fracture penetration according to the source radius of the fracture point i , to judge whether there may be a fracture penetration relationship between fracture points. Convert the data into a standard normal distribution with zero mean and unit variance through the Z-score normalization method to standardize the volume of the fracture points;

[0088] S16: When two adjacent fracture points \(i\) and \(j\) meet the fracture penetration conditions, calculate the fracture penetration possibility index \(BCI\) between the two adjacent fracture points ij , and calculate the total fracture penetration index \(BCI\) of fracture point \(i\) therefrom i , the formula is as follows:

[0089]

[0090] In the formula: \(Z(V\) i ) and \(Z(V\) j ) are the standardized volume values of the \(i\)-th fracture point and the \(j\)-th fracture point respectively; \(d\) ij is the Euclidean distance between the \(i\)-th fracture point and the \(j\)-th fracture point; \(I\) ij is an indicator function, and when \(d\) ij ≤ \(D\) i , its value is 1, indicating that the two points meet the fracture penetration conditions.

[0091] In an example of the present invention, in the step S20, as Figure 3 shown, the DBSCAN algorithm is used to perform density clustering on fractures, including the following steps:

[0092] S211: Multidimensional feature vector reading and Euclidean distance calculation: Read the multidimensional feature vectors (spatial coordinates, spatial orientation, volume, and energy) of each fracture point obtained by source mechanism inversion at each moment, and calculate the Euclidean distance between fracture points; among them, for each pair of fracture points \(P\) i = \((x\) i , \(y\) i , \(z\) i , \(nx\) i , \(ny\) i , \(nz\) i , \(V\) i , \(E\)i ) and P j =(x j , y j , z j , nx j , ny j , nz j , V j , E j ), and its Euclidean distance calculation formula is as follows:

[0093]

[0094] In the formula: n is the dimension of the feature vector, P i,k and P i,k are respectively the values of the k-th dimension in the multi-dimensional feature vectors of two fracture points, and w k is the weight of each feature (which can be adjusted according to the importance of different features);

[0095] S212: Determine the neighborhood distance and the minimum number of points: Select an appropriate neighborhood distance ∈ and the minimum number of points MinPts. ∈ determines the neighborhood radius of a certain fracture 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 ∈, calculate the neighborhood of each fracture point; if the number of points included in the neighborhood of a fracture point is greater than or equal to MinPts, then this fracture point is a core point; otherwise, it is a boundary point or a noise point;

[0096] S213: Cluster expansion and core point processing: Select an unlabeled core point as the initial point, mark it as the starting point of a new cluster, and assign a cluster number to this point; then, 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 the unlabeled core points for cluster expansion until all core points are processed;

[0097] That is to say, such as Figure 3As shown, for the core points, it is necessary to simultaneously determine whether each core point has a label and whether all core points have been read. If a core point has no label, its neighborhood needs to be expanded and a label is set. Otherwise, the next core point is read. If there are unread points among all the core points, it is necessary to continue to determine whether this core point has a label. Otherwise, all crack clusters are generated. Among them, after the core point expands its neighborhood and sets a label, it is also necessary to determine 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 determine whether the newly added points in the neighborhood are core points. Otherwise, keep the label unchanged and generate new crack clusters to form all crack clusters.

[0098] S214: Cluster number and clustering result check: According to the result of DBSCAN clustering, assign a unique crack surface number to each cluster, and each cluster represents an independent crack surface. If a fissure 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.

[0099] In an example of the present invention, in the step S20, potential fissure surfaces with high penetrability and expansion potential are identified by optimizing the combination of the fracture penetration possibility index BCI, and the B-spline surface fitting algorithm is applied to smoothly fit the fissure surfaces to obtain a reconstructed fissure surface image, including the following steps:

[0100] S221: Optimize and screen fissures based on BCI values: For each fissure cluster, calculate the fracture penetration possibility index BCI of each fissure therein, and further visualize the BCI value distribution of the fissures, set a suitable threshold, and select fissure points with higher BCI values as potential fissure surfaces.

[0101] S222: Reconstruct the time series image of the fissure surface: Read the fissure data obtained by clustering analysis at all times, combine the characteristic information of the spatial coordinates, BCI values, volumes, and normal directions of the fissures at each time, and perform fitting processing on the fissures at each time. Through the B-spline curve and triangular mesh division method, accurately reconstruct the fissure surface at each time, and finally integrate the reconstruction results of the fissure surfaces at all times to generate a complete time series image of the fissure surface.

[0102] In an example of the present invention, in the step S30, preprocess the data and divide the data set, including:

[0103] Convert the rupture penetration probability index of each fracture point on the fracture surface obtained by clustering fitting into the corresponding gray value, interpolate it onto a uniform 128×128×128 voxel grid, perform normalization processing on the interpolated image, and construct a spatio-temporal data set of fracture surface images. Use 80% of the data set as the training set, 10% as the validation set, and 10% as the test set.

[0104] In an example of the present invention, in the step S40, as Figure 4 and Figure 5 shown, the fracture surface image sequence prediction model includes a spatial encoder, a translator, and a spatial decoder.

[0105] The spatial encoder is configured to map the input high-dimensional fracture map data to a low-dimensional latent space, use three-dimensional convolution operations to capture spatial features in the three-dimensional fracture map data, and introduce a multi-head attention mechanism before convolution to capture spatio-temporal dependencies.

[0106] The translator is configured to use the Poolformer network as the translator of the prediction network, learn the spatial dependencies and temporal changes of the fracture surface image sequence from the latent space, take the output tensor of the encoder as the input of the Poolformer, stack multiple sequence frames along the time axis direction, then use the Poolformer to learn from the stacked multi-frame features, capture the inherent temporal evolution law in the fracture surface image sequence data, and finally take the output result as the input of the decoder.

[0107] The spatial decoder is configured to decode the information in the latent space into the predicted future fracture surface image, use three-dimensional transposed convolution operations to gradually restore the details of the image, and introduce a multi-head attention mechanism before transposed convolution to capture the spatio-temporal dependencies of the decoder features.

[0108] In an example of the present invention, the spatial encoder includes:

[0109] The first multi-head attention mechanism module, as Figure 6 shown, is configured to map the original features of the input fracture map to different feature spaces of query Q (query), key K (key), and value V (value), and pack them into matrices Q 1 、K 1 、V 1 , take Q 1 、K 1 、V 1 perform linear transformation with different weight matrices to obtain Q 1 ’、K 1 ’、V 1'; Suppose there are h attention heads, and each attention head has its own transformation matrices for Q, K, and V. For the i-th attention head, the transformation formulas for Q, K, and V are as follows:

[0110] Q′ i =QW Qi

[0111] K′ i =KW Ki

[0112] V i ′=VW Vi

[0113] (Q′ i ,K′ i ,V i ′)=(QW Qi ,KW Ki ,VW Vi )

[0114] Where: W Qi 、W Ki 、W Vi are the weight matrices of the i-th attention head respectively;

[0115] Then, for each attention head, calculate its attention value independently, and the formula is as follows:

[0116]

[0117] Where: head i is the attention value of the i-th attention head, Attention represents the attention function, softmax represents the activation function, and d k is the scaling factor of the vector dimension to avoid the problem of gradient disappearance caused by too large numerical values;

[0118] Finally, concatenate the outputs of all heads and obtain the final multi-head attention output through the final linear transformation. The formula is as follows:

[0119]

[0120] Where: is the output tensor calculated through the multi-head attention mechanism, MultiHead represents the multi-head attention output, Concat represents concatenation, and W O is the linear transformation matrix of the output;

[0121] The first local residual connection module. The first local residual connection module is set at the input and output ends of the first multi-head attention mechanism module, and is configured to maintain the transmission of spatial information. The formula is as follows:

[0122]

[0123] where: z' 0 is the feature tensor obtained through the residual connection, and z 0 is the initial feature tensor input into the multi-head attention mechanism, is the output tensor calculated through the multi-head attention mechanism;

[0124] The three-dimensional convolution module includes multiple cascaded convolution layers. The convolution layer includes: convolution, normalization operation, and non-linear transformation. The three-dimensional convolution module is configured to perform feature extraction on z' output by the first multi-head attention mechanism module after passing through the first local residual connection module, 0 where a downsampling operation is performed once every two convolution layers.

[0125] That is, to input z' output by the first multi-head attention mechanism module through the residual connection 0 into multiple convolution layers for feature extraction, including 6 cascaded convolutions, normalization operations, and non-linear transformations. The formula is as follows:

[0126] z i = SiLU(Norm3d(Conv3d(z i-1 ))) 1 ≤ i ≤ 6

[0127] where: SiLU is the sigmoid linear unit function for non-linear transformation, Norm3d is the three-dimensional normalization layer, Conv3d is the three-dimensional convolution operation, and z i represents the tensor input into the convolution layer, and the stride size of the convolution layer is set to 1;

[0128] A downsampling operation is performed once every two convolution layers, and its stride size is set to 2. The formula is as follows:

[0129] z i = MaxPool3d(z i , stride = 2)

[0130] where: MaxPool3d represents the three-dimensional max pooling operation.

[0131] In an example of the present invention, the spatial decoder includes:

[0132] The second multi-head attention mechanism module, as Figure 6 shown, is configured to map the output tensor w of the translator 0 to different feature spaces for query Q (query), key K (key), and value V (value), and respectively package them into matrices Q 2 , K 2 , V 2 , and Q2 , K 2 , V 2 is linearly transformed with different weight matrices to obtain Q 2 ’, K 2 ’, V 2 ’. Suppose there are h attention heads, and each attention head has its own transformation matrices for Q, K, and V. For the i-th attention head, the transformation formulas for Q, K, and V are as follows:

[0133] Q i ′ = QW Qi

[0134] K i ′ = KW Ki

[0135] V i ′ = VW Vi

[0136] (Q′ i , K′ i , V i ′) = (QW Qi , KW Ki , VW Vi )

[0137] In the formula: W Qi , W Ki , W Vi are the weight matrices of the i-th attention head respectively;

[0138] Next, for each attention head, calculate its attention value independently, and the formula is as follows:

[0139]

[0140] In the formula: head i is the attention value of the i-th attention head, Attention represents the attention function, softmax represents the activation function, d k is the scaling factor of the vector dimension to avoid the problem of gradient disappearance caused by too large numerical values;

[0141] Finally, concatenate the outputs of all heads and perform the final linear transformation to obtain the final multi-head attention output, and the formula is as follows:

[0142]

[0143] In the formula: is the final output tensor calculated through the multi-head attention mechanism, MultiHead represents the multi-head attention output, Concat represents concatenation, and W O is the linear transformation matrix of the output;

[0144] The second local residual connection module: The second local residual connection modules are respectively arranged at the input and output ends of the second multi-head attention mechanism module, and are configured to maintain the transmission of spatial information. The formula is as follows:

[0145]

[0146] In the formula: w′ 0 is the feature tensor obtained through the residual connection, w 0 is the output tensor of the translator input into the multi-head attention mechanism, is the final output tensor calculated through the multi-head attention mechanism;

[0147] The 3D transposed convolution module includes multiple transposed convolution layers connected in series. The transposed convolution layer includes: transposed convolution, normalization operation, and non-linear transformation. The 3D transposed convolution module is configured to perform feature extraction and spatial resolution restoration on the output w′ of the second multi-head attention mechanism module passing through the second local residual connection module. Among them, for every two transposed convolution layers, an upsampling operation is performed using PixelShuffle. 0 That is, to input the output w′ of the second multi-head attention mechanism module with residual connection

[0148] into multiple transposed convolution layers for feature extraction and spatial resolution restoration, including 6 transposed convolutions, normalization operations, and non-linear transformations connected in series. The formula is as follows: 0 w

[0149] w k = SiLU(Norm3d(UnConv3d(w k-1 ))) 1 ≤ k ≤ 6

[0150] In the formula: SiLU is the sigmoid linear unit function for non-linear transformation, Norm3d is the 3D normalization layer, UnConv3d is the 3D transposed convolution operation, w i represents the tensor input into the transposed convolution layer, and the stride size of the transposed convolution layer is set to 1;

[0151] For every two transposed convolution layers, an upsampling operation is performed using PixelShuffle, and its stride size is set to 2. The formula is as follows:

[0152] w k = PixelShuffle3d(w k , stride = 2)

[0153] In an example of the present invention, the step S50 specifically includes the following steps:

[0154] S51: Train the crack surface image sequence prediction model using the training set, adjust the batchsize, learning rate, optimizer, and number of epochs, and adopt the L2 loss function and the motion perception loss function L m Construct the total loss function L to supervise the model training; the expression of the loss function is as follows:

[0155]

[0156] L = λ 1 L2 + λ 2 L m

[0157] In the formula: N is the length of the prediction sequence, Y i and respectively represent the true value and the predicted value at the i-th time step, COF represents the function for calculating the dense optical flow to calculate the crack movement change between adjacent two frames, λ 1 and λ 2 are respectively the L2 loss coefficient and the motion perception loss coefficient;

[0158] S52: Input the test set into the trained crack surface image sequence prediction model to obtain the predicted crack surface image sequence, and use the evaluation metrics of MSE, SSIM, and PSNR to evaluate the prediction results. The evaluation metrics are as follows:

[0159]

[0160] In the formula, Y i and respectively represent the true value and the predicted value at the i-th time step, μ Y and respectively represent the average values of Y i and σ Y and respectively represent the standard deviations of Y i and , represents the covariance of Y i and c 1 and c 2 are constants, I MAX represents the maximum pixel value of the image.

[0161] In an example of the present invention, in the step S52, inputting the test set into the trained crack surface image sequence prediction model to obtain the predicted crack surface image sequence specifically includes the following steps:

[0162] Input the crack surface image sequence X ∈ R into the prediction model T×C×H×W×D, where T represents the time step, C represents the number of channels of each image, and H, W, and D represent the height, width, and depth of each image, respectively.

[0163] A set of dynamic sequences is formed based on the change X at T time steps. The prediction of the fracture surface image sequence predicts the future T' frames based on the given T-frame images, and its modeling formula is as follows:

[0164]

[0165] In the formula: X T×C×H×W×D is the input fracture surface image sequence, is the output predicted fracture surface image sequence, Y T′×C×H×W×D is the real fracture surface image sequence, and Θ represents a series of parameters to be optimized;

[0166] An engineering rock mass fracture surface dynamic expansion prediction system according to the second aspect of the present invention includes:

[0167] A parameter acquisition and calculation module, configured to obtain the original acoustic wave data collected by the microseismic monitoring system through extraction, perform positioning calculation and focal mechanism inversion on it, obtain the core parameters of the coal and rock fracture source directly related to the fracture at each moment, and calculate the fracture penetration possibility index BCI of the fracture at each moment in combination with the standardized fracture volume, source radius, and spatial distance;

[0168] A fracture surface image reconstruction module, configured to calculate the similarity between fractures at each moment according to multi-dimensional features such as the spatial distance, orientation, volume, and energy of the fractures, perform density clustering on the fractures using the DBSCAN clustering algorithm, optimize and identify potential fracture surfaces with high connectivity and expansion potential in combination with the fracture penetration possibility index BCI, and apply the B-spline surface fitting algorithm to smooth and fit the fracture surface to obtain the reconstructed fracture surface image;

[0169] A data set processing and division module, configured to arrange the reconstructed fracture surface images in chronological order to form a fracture surface time series image; after preprocessing, construct a fracture surface image spatio-temporal data set and divide it into a training set, a validation set, and a test set;

[0170] A prediction model construction module, configured to construct a fracture surface image sequence prediction model using an improved SimVP spatio-temporal neural network, including introducing a multi-head self-attention mechanism and local residual connections in the encoder and decoder, and replacing the translator with a Poolformer network;

[0171] The fracture surface image prediction module is configured to train the fracture surface image sequence prediction model by inputting the training set, and use the validation set to test and evaluate the trained prediction model, and continuously adjust the parameters for training. After training, the fracture surface sequence images of the test set are input into the trained prediction network to obtain the predicted fracture surface image sequence, and the corresponding evaluation indexes are used to systematically evaluate the prediction results.

[0172] This prediction system obtains the core parameters of the focal mechanism through microseismic monitoring inversion, can accurately capture the source characteristics of fractures, and uses them as the input data of the DBSCAN clustering algorithm, significantly improving the accuracy and reliability of the reconstruction of the fracture surface time series images, and providing more scientific technical support for the further study of the dynamic expansion prediction of fractures.

[0173] Different from traditional time series prediction models, the SimVP network adopted by this prediction system is a video prediction network completely based on the CNN architecture, without complex recurrent structures or additional modules, avoiding the introduction of too many techniques and strategies, thus achieving efficient prediction of the fracture expansion process while keeping the network structure simple.

[0174] When constructing the fracture surface image sequence prediction model, this prediction system introduces the multi-head attention mechanism in the encoder and decoder, effectively extracts the high-dimensional features of the fracture surface image sequence data, enables the model to maintain the spatial information of the sequence while learning the temporal correlation, thus improving the accuracy and reliability of the fracture surface prediction.

[0175] Through the comprehensive application of focal mechanism inversion, DBSCAN clustering algorithm and spatio-temporal neural network prediction model, this prediction system fully integrates the advantages of multiple technologies and effectively captures the spatio-temporal correlation in the fracture expansion process. And the prediction model can deeply mine the time series features and spatial distribution patterns in the fracture expansion process, significantly improving the prediction accuracy of the dynamic evolution of the fracture surface, providing a basis for optimizing the design and implementation of hydraulic fracturing.

[0176] In the above text, the exemplary implementation manners of the engineering rock mass fracture surface dynamic expansion prediction method and system proposed by the present invention are 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 of the technical features and structures proposed by the present invention can be made, 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 method for predicting the dynamic expansion of fracture surfaces in engineering rock mass, characterized in that: The steps include: S10: Obtain the original acoustic wave data and perform positioning calculation and focal mechanism inversion on it, obtain the core parameters of the coal rock fracture source directly related to the fracture at each moment, and calculate the fracture penetration possibility index BCI of the fracture at each moment in combination with the standardized fracture volume, focal radius and spatial distance; S20: Calculate the similarity between cracks at each moment based on the multidimensional characteristics of the cracks, and use the DBSCAN clustering algorithm to cluster the cracks. Combined with the fracture penetration probability index BCI optimization, identify potential crack surfaces with high penetration and expansion potential. Apply the B-spline surface fitting algorithm to smoothly fit the crack surface to obtain the reconstructed crack surface image. S30: arranging the reconstructed fracture surface images in chronological order to form fracture surface time series images; After preprocessing, a spatiotemporal dataset of fracture surface images is constructed and divided into training set, validation set and test set; S40: Construction of fracture surface image sequence prediction model using improved SimVP spatiotemporal neural network; S50: inputting the training set into the fracture surface image sequence prediction model for training, using the validation set to test and evaluate the trained prediction model, and continuously adjusting the training parameters; After the training is completed, the fracture surface sequence images of the test set are input into the trained prediction network to obtain the predicted fracture surface image sequence, and the prediction results are systematically evaluated using the corresponding evaluation indicators.

2. The method for predicting dynamic expansion of engineering rock mass fracture surfaces according to claim 1, characterized in that: In the step S10, the original acoustic wave data is subjected to positioning calculation and focal mechanism inversion, the focal mechanism core parameters directly related to the rupture at each moment are obtained, and the rupture penetration probability index BCI of the fissure at each moment is calculated, 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 sound 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; 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: quantitatively invert and calculate the core parameters of the coal-rock fracture source according to the calculated moment tensor components, wherein the core parameters of the coal-rock fracture source include fracture spatial orientation, fracture volume and fracture energy; S15: Determine the critical distance D of the fracture penetration according to the earthquake source radius of the fracture point i , the data were transformed into a standard normal distribution with zero mean and unit variance by the Z-score standardization method; S16: When two adjacent crack points i and j meet the crack penetration condition, calculate the crack penetration probability index BCI between the two adjacent crack points ij , and the total fracture penetration index BCI of crack point i is calculated from this i .

3. The method for predicting dynamic expansion of engineering rock mass fracture surfaces according to claim 1, characterized in that: In step S20, the DBSCAN algorithm is used to perform density clustering on the cracks, including the following steps: S211: Reading of multi-dimensional feature vectors and calculation of Euclidean distance: Read the multi-dimensional feature vector of each fracture point obtained by focal mechanism inversion at each moment, and calculate the Euclidean distance between fracture points; S212: Determine neighborhood distance and minimum number of points: Select appropriate neighborhood distance ∈ and minimum number of points MinPts, and calculate the neighborhood of each crack point based on Euclidean distance and 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; S213: Cluster expansion and core point processing: select one of the unmarked core points as an initial point, mark it as the starting point of a new cluster, and assign a cluster number to the point; Then, the neighborhood of the core point is expanded, and all points in the neighborhood are classified 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 the unlabeled core points for cluster expansion until all core points have been processed. S214: Check cluster number and clustering results: According to the results of DBSCAN clustering, 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 noise point is numbered -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.

4. The method for predicting dynamic expansion of engineering rock mass fracture surfaces according to claim 1, characterized in that: In step S20, the potential fracture surface with high penetration and expansion potential is identified by optimizing the fracture penetration probability index BCI, and the fracture surface is smoothly fitted by applying the B-spline surface fitting algorithm to obtain a reconstructed fracture surface image, including the following steps: S221: Optimize and screen cracks based on BCI values: For each crack cluster, calculate the fracture penetration probability index BCI of each crack in it, and further visualize the BCI value distribution of the cracks, set a suitable threshold, and select the crack points with higher BCI values ​​as potential crack surfaces; S222: Reconstruct the time series image of the fracture surface: read the fracture data obtained by cluster analysis at all times, and fit the fracture at each moment based on the characteristic information of the spatial coordinates, BCI value, volume and normal direction of the fracture at each moment; accurately reconstruct the fracture surface at each moment through B-spline curve and triangular mesh division method, and finally integrate the reconstruction results of the fracture surface at all moments to generate a complete time series image of the fracture surface.

5. The method for predicting the dynamic expansion of fracture surfaces in engineering rock mass according to claim 1, characterized in that: In step S40, the crack surface image sequence prediction model includes a spatial encoder, a translator and a spatial decoder. The spatial encoder is configured to map the input high-dimensional crack map data to a low-dimensional latent space, use a three-dimensional convolution operation to capture the spatial features in the three-dimensional crack map data, and introduce a multi-head attention mechanism before the convolution to capture the spatiotemporal dependency; The translator is configured to use a Poolformer network as a translator of a prediction network to learn the spatial dependency and temporal variation of the crack surface image sequence from the latent space, use the output tensor of the encoder as the input of the Poolformer, and then stack multiple sequence frames along the time axis, and then use the Poolformer to learn from the stacked multi-frame features to capture the inherent temporal evolution law of the crack surface image sequence data, and finally use the output result as the input of the decoder; The spatial decoder is configured to decode the information in the latent space into a predicted future crack surface image, gradually restore the details of the image using a three-dimensional deconvolution operation, and introduce a multi-head attention mechanism before deconvolution to capture the spatiotemporal dependencies of the decoder features.

6. The method for predicting the dynamic expansion of fracture surfaces in engineering rock mass according to claim 5, characterized in that: The spatial encoder comprises: The first multi-head attention mechanism module is configured to map the original features of the input crack map to different feature space queries Q, keys K and values ​​V, and pack them into matrices Q1, K1, and V1 respectively, and obtain Q1', K1', and V1' by linearly transforming Q1, K1, and V1 with different weight matrices; then, for each attention head, its attention value is calculated independently; finally, the outputs of all heads are spliced ​​and the final multi-head attention output is obtained through the final linear transformation; A first local residual connection module is provided at the input and output ends of the first multi-head attention mechanism module, respectively, and is configured to maintain the transmission of spatial information; A three-dimensional convolution module includes multiple convolution layers connected in series, wherein the convolution layers include: convolution, normalization operation and nonlinear transformation. The three-dimensional convolution module is configured to perform feature extraction on z′0 output by the first multi-head attention mechanism module after the first local residual connection module, wherein a downsampling operation is performed on every two convolution layers.

7. The method for predicting the dynamic expansion of fracture surfaces in engineering rock mass according to claim 5, characterized in that: The spatial decoder comprises: The second multi-head attention mechanism module is configured to map the translator's output tensor w0 to different feature space queries Q, keys K and values ​​V, and pack them into matrices Q2, K2, and V2 respectively, and obtain Q2', K2', and V2' by linearly transforming Q2, K2, and V2 with different weight matrices; then, for each attention head, its attention value is calculated independently; finally, the outputs of all heads are concatenated and subjected to the final linear transformation to obtain the final multi-head attention output; A second local residual connection module is provided at the input and output ends of the second multi-head attention mechanism module, respectively, and is configured to maintain the transmission of spatial information; A three-dimensional deconvolution module includes multiple serially connected deconvolution layers, wherein the deconvolution layers include: deconvolution, normalization operation and nonlinear transformation. The three-dimensional deconvolution module is configured to extract w′0 row features and restore spatial resolution of the output of the second multi-head attention mechanism module after the second local residual connection module, wherein an upsampling operation is performed once for every two deconvolution layers using PixelShuffle.

8. The method for predicting the dynamic expansion of fracture surfaces in engineering rock mass according to claim 1, characterized in that: The step S50 specifically includes the following steps: S51: Train the crack surface image sequence prediction model through the training set, adjust the batch size, learning rate, optimizer and epoch number, and use the L2 loss function and motion perception loss function L m Construct a total loss function L to supervise the model training; S52: Input the test set into the trained fracture surface image sequence prediction model to obtain the predicted fracture surface image sequence, and use MSE, SSIM, and PSNR evaluation indicators to evaluate the prediction results. The evaluation indicators are as follows: Where Y i and Represent the true value and predicted value of the i-th time step, μ Y and Respectively represent Y i and The average value, σ Y and Respectively represent Y i and The standard deviation of Represents Y i and covariance, c1 and c2 are constants, I MAX Indicates the maximum pixel value of the image.

9. The method for predicting dynamic expansion of fracture surfaces in engineering rock mass according to claim 8, characterized in that: In step S52, the test set is input into the trained fracture surface image sequence prediction model to obtain a predicted fracture surface image sequence, which specifically includes the following steps: Input the crack surface image sequence X∈R into the prediction model T×C×H×W×D , where T represents the time step, C represents the number of channels of each image, H, W, D represent the height, width and depth of each image respectively; According to the changes X in T time steps, a set of dynamic sequences is formed. The crack surface image sequence prediction predicts the future T' frames based on the given T frame images. The modeling formula is as follows: Where: X T×C×H×W×D is the input crack surface image sequence, is the output predicted crack surface image sequence, Y T′×C×H×W×D is a real crack surface image sequence, and Θ represents a series of parameters that need to be optimized.

10. A dynamic expansion prediction system for engineering rock mass fracture surface, characterized in that: include: The parameter acquisition and calculation module is configured to obtain the original acoustic wave data and perform positioning calculation and focal mechanism inversion on it, so as to obtain the core parameters of the coal and rock fracture source directly related to the fracture at each moment, and calculate the fracture penetration possibility index BCI of the fracture at each moment in combination with the standardized fracture volume, focal radius and spatial distance; A fracture surface image reconstruction module is configured to calculate the similarity between fractures at each moment according to the multi-dimensional characteristics of the fractures, and to perform density clustering on the fractures using the DBSCAN clustering algorithm, and to identify potential fracture surfaces with high penetration and expansion potential by combining the fracture penetration probability index BCI optimization, and to perform smooth fitting on the fracture surface using the B-spline surface fitting algorithm to obtain a reconstructed fracture surface image; The data set processing and partitioning module is configured to arrange the reconstructed fracture surface images in chronological order to form fracture surface time series images; after preprocessing, construct a fracture surface image spatiotemporal data set and divide it into a training set, a validation set and a test set; A prediction model building module, configured to build a crack surface image sequence prediction model using an improved SimVP spatiotemporal neural network; A fracture surface image prediction module is configured to input a training set into a fracture surface image sequence prediction model for training, use a validation set to test and evaluate the trained prediction model, and continuously adjust parameter training; After the training is completed, the fracture surface sequence images of the test set are input into the trained prediction network to obtain the predicted fracture surface image sequence, and the prediction results are systematically evaluated using the corresponding evaluation indicators.

Images