A geological data pseudo-three-dimensional conversion model construction method and system

Abstract

Compared with the conventional pseudo-three-dimensional interpolation method, the geological data pseudo-three-dimensional conversion model construction method and system overcomes the non-uniform distribution characteristics of downhole data, increases the adaptability of geological data three-dimensional technology, fuses historical geological information including geological exposure and three-dimensional seismic, also fuses dynamic exploration and detection results as the mine is excavated, based on the geological data characteristics, applies the geological constraint interpolation and the generated adversarial grid method, and obtains the pseudo-three-dimensional geological data body with higher precision and stronger reliability in the working face scale. By using the stage optimization and conditional generation method under the geological constraint condition, on the basis of the ground seismic and downhole two-dimensional detection, the uncertainty of the geological data is significantly reduced, the spatial geological information exposed in other directions can be comprehensively used, a multiple loss function is constructed, the high and low frequency data body has the ability to reflect the three-dimensional geological details, and is suitable for large-scale use and popularization in industry.

Description

Technical Field

[0001] This invention belongs to the field of geophysical exploration technology, specifically relating to a method and system for constructing a pseudo-three-dimensional conversion model of geological data. Background Technology

[0002] Geological exploration during tunneling is one of the most important geological assurance methods in intelligent tunneling. It uses reflected acoustic wave technology for advanced detection, enabling dynamic identification of geological structures within a certain range ahead of the tunnel. Due to the limited working conditions in coal mines, this method employs a two-dimensional linear observation system, which has the advantage of close proximity between the seismic source and the detection target. It can collect reflected vibration signals generated during tunneling in real time, obtaining two-dimensional geological structure detection results in the horizontal direction ahead of the tunnel. Due to the limitations of the underground working space, the geological exploration data is a regional, horizontal strip-shaped two-dimensional result, making it difficult to reconstruct a continuous, true three-dimensional image. The description of geological phenomena ahead of the tunnel is limited to the horizontal direction. The geological data from the underground working face also cannot meet the prerequisite of a pseudo-three-dimensional uniform distribution. The pseudo-three-dimensional data volume generated by conventional methods has a narrow frequency band and weak geological identification ability.

[0003] Existing methods have the following limitations: First, in unknown areas lacking constraints from real geological information, interpolation uncertainty increases significantly, and the sparse effective data points between survey lines result in low accuracy and reliability of the generated pseudo-3D data volume. Second, traditional pseudo-3D processing methods are limited by static data sources, only providing static pseudo-3D volumes, which is insufficient to meet the requirements of dynamic 3D analysis ahead of tunneling in underground coal mine operations. Pseudo-3D generation of dynamic 2D results from tunneling exploration has the potential to provide dynamic multi-angle analysis of the tunneling geological environment.

[0004] Machine learning is one of the important methods for realizing the three-dimensionalization of data at present, but existing methods have limitations: conventional generative adversarial networks (GANs), as a general image super-resolution technology, have difficulty in generating results that conform to the deposition rules and have insufficient accuracy. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for constructing a pseudo-3D conversion model of geological data, so as to solve the problem of insufficient accuracy in generating pseudo-3D address data in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A method for constructing a pseudo-3D conversion model of geological data includes the following steps: Step 1: For the coal mine working face to be processed, delineate the three-dimensional geometric framework that can cover the coal mine working face; Based on the geological data of the coal mine working face, three-dimensional amplitude data volume, two-dimensional excavation data and waveform of marker strata were obtained respectively; The three-dimensional amplitude data volume includes multiple seismic traces distributed in three-dimensional space, with multiple sampling points distributed on each seismic trace, and each sampling point corresponding to an amplitude value. Step 2: Divide the three-dimensional geometric framework into multiple three-dimensional meshes of the same size to form the initial geometric model; For the three-dimensional amplitude data volume, amplitude balancing is performed. For any sampling point in the amplitude-balanced three-dimensional amplitude data volume, its normal vector is calculated. The tilt angle is obtained based on the normal vector, and then the tilt angle value is filled into the corresponding sampling point to obtain the tilt angle attribute volume. Traverse all sampling points and average the dip angle attribute volume to the initial geometric model to form a three-dimensional dip angle attribute model covering the coal mine working face; Step 3: Based on the waveform of the marker layer, trace the marker layer surface in the initial geometric model; Next, using the marker layer surface as an index, the dip angle attribute values ​​of the marker layer are extracted from the three-dimensional dip angle attribute model; The horizontal projection correction of the two-dimensional excavation data is performed based on the dip angle attribute value of the marker layer to obtain the horizontal component of the two-dimensional excavation data; Step 4: Based on the three-dimensional vector of the seismic trace of the three-dimensional amplitude data volume after amplitude balance and the horizontal component of the two-dimensional excavation data, establish the interpolation objective function to calculate the pseudo-three-dimensional data, as shown in the following formula;

[0007] in, This represents the pseudo-3D data to be solved; Represents the spatial gradient operator for the tilt field; Represents a double integral; This represents the data fitting weights, ranging from 0.01 to 10. Represents the horizontal component of two-dimensional in-situ data; A three-dimensional seismic vector representing the seismic data after amplitude equilibrium; Indicates the index of the sampling point; The dip field spatial gradient operator is represented as follows:

[0008] in, Indicates the partial derivative sign; It represents the horizontal coordinates, specifically the horizontal coordinates along the horizontal axis within the coal mine working face; This represents the horizontal coordinates, specifically the inner edge of the coal mine working face, dip direction, and direction. Horizontal coordinates on orthogonal horizontal axes; Indicates the dip angle of the strata; Using the minimization of the interpolation objective function as the baseline condition, three-dimensional interpolation is performed for each three-dimensional grid to obtain low-frequency pseudo-three-dimensional data. Step 5: Construct a pseudo-3D conversion model of geological data, including an encoder, decoder, and discriminator; The encoder includes multiple convolutional layers; The decoder is a transposed convolutional structure symmetric to the encoder; The discriminator described above employs a multi-layer convolutional structure and a binary cross-entropy loss function. Step 6: Use low-frequency pseudo-3D data as input to the geological data pseudo-3D conversion model and dip angle attribute model as constraint condition to train the geological data pseudo-3D conversion model and obtain the trained geological data pseudo-3D conversion model.

[0009] The present invention also has the following features: Furthermore, step 6 specifically includes the following sub-steps: Step 61: The encoder extracts features from the low-frequency pseudo-3D data layer by layer through multiple convolutional layers and outputs a feature image. Step 62: The decoder uses the tilt attribute model as a constraint and upsamples the feature image layer by layer through transposed convolution. After upsampling, it splices the data through skip connection layers to output high-frequency pseudo-3D data. Step 63: Input the three-dimensional amplitude data volume and high-frequency pseudo-three-dimensional data into the discriminator. The discriminator uses a convolutional layer and a binary cross-entropy loss function to compress the feature dimension to 1×1×128 for classification. Step 64: Establish a total loss function that includes data distribution loss, structural consistency loss, and spectrum matching loss. Calculate the total loss value and the loss function. As shown in the following formula:

[0010] in, Represents the data distribution loss function; Represents the structural consistency loss function; Represents the spectrum matching loss function; Step 65: With the goal of minimizing the total loss function value, train the pseudo-3D transformation model of geological data until the model converges, and obtain the trained pseudo-3D transformation model of geological data.

[0011] Furthermore, in step 63, the data distribution loss function is as follows:

[0012] in, Represents the mathematical expectation operator; This represents the probability that the discriminator will distinguish the input data; Represents a three-dimensional amplitude data volume; This indicates that the decoder takes low-frequency pseudo-3D data as input and outputs high-frequency pseudo-3D data. This represents low-frequency pseudo-3D data; The structural consistency loss function is as follows:

[0013] in, Represents the spatial gradient operator for the tilt field; This represents the high-frequency pseudo-3D data output by the decoder, i.e. ; Represents the L1 norm; The spectrum matching loss function is as follows:

[0014] in, Indicates the weighting coefficient; Indicates Fourier transform; Represents the L2 norm; Furthermore, step 2 includes the following sub-steps: Step 21: Divide the three-dimensional geometric framework into multiple three-dimensional meshes of the same size to form an initial geometric model; Step 22: Apply a normalization method to the amplitude values ​​of each seismic trace in the 3D amplitude data volume to perform amplitude balancing, obtaining the 3D amplitude data volume after amplitude balancing; specifically, the normalization coefficient is as follows:

[0015] in, Represents the normalization coefficient; Indicates the longitudinal position of the sampling point; Indicates the amplitude of the sampling points on the seismic trace; Indicates the root mean square amplitude; Specifically,

[0016] in, Indicates the length of the sliding window; Indicates the current sampling point; Step 23: For any sampling point in the three-dimensional amplitude data volume after amplitude balance, calculate its normal vector using the gradient method, determine the tilt angle based on the normal vector, fill the tilt angle value into the corresponding sampling point, and obtain the tilt angle attribute volume; Step 24: Repeat step 23 until every sampling point has been traversed; average the dip angle attribute volume onto the initial geometric model to form a three-dimensional dip angle attribute model covering the coal mine working face. .

[0017] A pseudo-3D conversion system for geological data includes a data acquisition and preprocessing module, a geological geometric modeling module, a spatial constraint correction module, a differential variable calculation module, a deep learning module, and a dynamic fusion output module; The data acquisition and preprocessing module is used for: Collect geological data from coal mine working faces; Delineate a three-dimensional geometric framework that can cover the coal mine working face; Amplitude balancing processing is performed on the three-dimensional amplitude data volume; The geological geometric modeling module is used for: The three-dimensional geometric framework is divided into multiple three-dimensional meshes of the same size to form an initial geometric model; Calculate the normal vector of the sampling point, obtain the dip angle based on the normal vector, fill the dip angle value into the corresponding sampling point to obtain the dip angle attribute volume; average the dip angle attribute volume to the initial geometric model to form a three-dimensional dip angle attribute model covering the coal mine working face; The aforementioned spatial constraint correction module is used for: Based on the waveform of the marker layer, trace the marker layer surface in the initial geometric model; Using the marker layer surface as an index, extract the marker layer dip angle attribute value from the three-dimensional dip angle attribute model; The horizontal projection correction of the two-dimensional excavation data is performed based on the dip angle attribute value of the marker layer to obtain the horizontal component of the two-dimensional excavation data. The variational interpolation calculation module is used for: Configure the interpolation objective function with minimizing the value of the interpolation objective function as the baseline condition, and perform three-dimensional interpolation for each three-dimensional grid to obtain low-frequency pseudo-three-dimensional data; The deep learning generation module has a built-in trained geological data pseudo-3D conversion model, which generates high-frequency pseudo-3D data based on low-frequency pseudo-3D data and a 3D dip angle attribute model. The dynamic fusion output module is used to fuse low-frequency pseudo-3D data and high-frequency pseudo-3D data to output a pseudo-3D geological data volume.

[0018] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of methods 1 to 4 above.

[0019] Compared with the prior art, the present invention has the following technical effects: (I) Compared with conventional pseudo-3D interpolation methods, the geological data pseudo-3D conversion model construction method and system of the present invention overcomes the non-uniform distribution characteristics of downhole data, increases the adaptability of geological data 3D conversion technology, integrates historical geological information including geological exposure and 3D seismic data, and also integrates dynamic on-the-go exploration results. Based on the characteristics of geological data, the method of geological constraint interpolation and generation of adversarial mesh is applied to obtain a pseudo-3D geological data volume with higher accuracy and stronger credibility within the working face scale.

[0020] (II) This invention utilizes a phased optimization and condition generation method under geological constraints to significantly reduce the uncertainty of geological data based on ground seismic and downhole two-dimensional exploration. It can integrate spatial geological information revealed from other directions, construct multiple loss functions, and fuse high- and low-frequency data volumes to reflect three-dimensional geological details, making it suitable for large-scale industrial use and promotion. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, all methods in this invention are methods known in the prior art.

[0022] The following are specific embodiments of the present invention. It should be noted that the present invention is not limited to the following specific embodiments. All equivalent modifications made based on the technical solutions of this application fall within the protection scope of the present invention.

[0023] A method for constructing a pseudo-3D conversion model of geological data includes the following steps: Step 1: For the coal mine working face to be processed, delineate the three-dimensional geometric framework that can cover the coal mine working face; Based on the geological data of the coal mine working face, three-dimensional amplitude data volume, two-dimensional excavation data and waveform of marker strata were obtained respectively; The three-dimensional amplitude data volume includes multiple seismic traces, each with multiple sampling points, and each sampling point corresponds to an amplitude value. Two-dimensional data acquired during excavation are grayscale images of seismic profiles collected along the excavation direction, including reflection amplitude values ​​at corresponding horizontal positions; Specifically, marker stratigraphy is a well-known concept in this field, referring to rock strata or interfaces that have distinct lithological characteristics, stable distribution, are easily identifiable, and have a certain thickness. Examples include mudstone on the roof of a coal seam, specific limestone layers, or the top / bottom interface of the coal seam itself.

[0024] A typical marker stratum is either the roof or floor of the main coal seam, or a stable interbedded rock layer within the coal seam.

[0025] Step 2: Divide the three-dimensional geometric framework into multiple three-dimensional meshes of the same size to form the initial geometric model; When dividing the three-dimensional mesh, the precision of the mechanical control of the intelligent tunneling equipment is referenced to divide the geometric framework model of the working face and form a three-dimensional mesh.

[0026] This embodiment provides a more specific implementation method, wherein the horizontal spacing of the three-dimensional element mesh is no greater than 1 / 2 of the size of the seismic element, and the vertical spacing is set to less than 0.2m; For the three-dimensional amplitude data volume, amplitude balancing processing is performed. For any sampling point in the amplitude-balanced three-dimensional amplitude data volume, its normal vector is calculated. The tilt angle is obtained based on the normal vector. The tilt angle value is filled into the corresponding sampling point to obtain the tilt angle attribute volume. Traverse all sampling points and average the dip angle attribute volume to the initial geometric model to form a three-dimensional dip angle attribute model covering the coal mine working face; Specifically, step 2 includes the following sub-steps: Step 21: Divide the three-dimensional geometric framework into multiple three-dimensional meshes of the same size to form an initial geometric model; Step 22: Apply a normalization method to the amplitude values ​​of each seismic trace in the 3D amplitude data volume to perform amplitude balancing, obtaining the 3D amplitude data volume after amplitude balancing; specifically, the normalization coefficient is as follows:

[0027] in, Represents the normalization coefficient; Indicates the longitudinal position of the sampling point; Indicates the amplitude of the sampling points on the seismic trace; Indicates the root mean square amplitude; Specifically,

[0028] in, Indicates the length of the sliding window; Indicates the current sampling point; Step 23: For any sampling point in the three-dimensional amplitude data volume after amplitude balance, calculate its normal vector using the gradient method, determine the tilt angle based on the normal vector, fill the tilt angle value into the corresponding sampling point, and obtain the tilt angle attribute volume; In this step, determining the tilt angle based on the normal vector is a well-known method in the art. This embodiment provides another specific implementation method: Since a unit normal vector n=(nx,ny,nz) has been calculated for each sampling point, representing the normal direction of the stratum at that point, the dip angle is the angle between the stratum interface and the horizontal plane, which is calculated from the angle between the normal vector and the vertical direction (z-axis). If the normal vector is upward, the dip angle is between 0° and 90°; if the normal vector is downward, the absolute value is taken.

[0029] In the three-dimensional mesh, the calculated tilt angle value is stored in an array cell at the same position as the original sampling point, forming a tilt angle attribute volume with the same dimensions as the three-dimensional amplitude data volume.

[0030] Step 24: Repeat step 23 until every sampling point has been traversed; average the dip angle attribute volume onto the initial geometric model to form a three-dimensional dip angle attribute model covering the coal mine working face. .

[0031] Step 3: Based on the waveform of the marker layer of the coal mine working face, trace the marker layer surface in the initial geometric model; using the marker layer surface as an index, extract the dip angle attribute value of the marker layer in the three-dimensional dip angle attribute model; This method is well-known in the art, and this embodiment provides another specific implementation method: The marker layer surface is a two-dimensional mesh surface that records the depth or time position of the marker layer at each horizontal coordinate. Traversing each point on the surface, using its corresponding depth as an index, the three-dimensional tilt angle attribute model is then used. Find the inclination angle value at that location. If the depth happens to fall within... If the node is on a grid, its tilt angle value is read directly; if it falls between two grid layers, the tilt angle value is obtained through linear interpolation. This ultimately yields the tilt angle distribution along the marker layer surface, where each location corresponds to a tilt angle value.

[0032] The horizontal projection correction of the two-dimensional tunneling data of the coal mine working face is performed based on the dip angle attribute value of the marker strata to obtain the horizontal component of the two-dimensional tunneling data; In this embodiment, specifically from the tilt angle attribute model In the process, the dip angle of the strata corresponding to each horizontal position is extracted along the surface of the marker layer. The horizontal projection correction is performed on the two-dimensional excavation data of the coal mine working face, and the corrected horizontal component is used for subsequent interpolation processing.

[0033] Step 4: Based on the three-dimensional vector of the seismic trace of the three-dimensional amplitude data volume after amplitude balance and the horizontal component of the two-dimensional excavation data, establish the interpolation objective function as follows;

[0034] in, This represents the pseudo-3D data to be solved; Represents the spatial gradient operator for the tilt field; Represents a double integral; This represents the data fitting weights, ranging from 0.01 to 10. Represents the horizontal component of two-dimensional in-situ data; A three-dimensional seismic vector representing the seismic data after amplitude equilibrium; Indicates the index of the sampling point; The dip field spatial gradient operator is represented as follows:

[0035] in, Indicates the sign of the partial derivative; It represents the horizontal coordinates, specifically the horizontal coordinates along the horizontal axis within the coal mine working face; This represents the horizontal coordinates, specifically the inner edge of the coal mine working face, dip direction, and direction. Horizontal coordinates on orthogonal horizontal axes; Indicates the dip angle of the strata; Using the minimization of the interpolation objective function as the baseline condition, three-dimensional interpolation is performed for each three-dimensional grid to obtain low-frequency pseudo-three-dimensional data. The following provides a further explanation of "low-frequency pseudo-3D data": In geological data, "low frequency" is a well-known concept, referring to macroscopic structural information that changes slowly and has strong spatial continuity. It mainly includes the overall undulation of strata (such as anticlines and synclines) and the spatial distribution trend of marker strata. In contrast, "high frequency" refers to local details that change drastically and have a small scale, such as faults, fractures, thin interbedded layers, and textures.

[0036] In the method of this embodiment, the interpolation objective function is constrained by the tilt field gradient operator, and the interpolation result tends to be smoother and more continuous. Therefore, the output is a background field without local details, i.e., a "low-frequency" component.

[0037] Pseudo-3D is a well-known concept distinct from true 3D. True 3D refers to a data volume with complete xyz spatial continuity, directly acquired through an actual 3D observation system. Pseudo-3D, on the other hand, is a data volume obtained by filling a 3D grid space with two-dimensional linear data (such as the two-dimensional excavation data in this embodiment) using various optional methods (interpolation fusion in this embodiment).

[0038] In this embodiment, the low-frequency pseudo-3D data has a spatial range of the 3D grid obtained in the above steps. By using the interpolation objective function and the dip field as a constraint, the 2D excavation data is extended to the 3D space along the stratum direction.

[0039] Step 5: Construct a pseudo-3D conversion model of geological data, including an encoder, decoder, and discriminator; The encoder specifically includes 8 convolutional layers; The decoder is a transposed convolutional structure that is symmetric to the encoder; The discriminator employs a multi-layer convolutional structure and a binary cross-entropy loss function; Step 6: Use low-frequency pseudo-3D data as input to the geological data pseudo-3D conversion model, and use the dip angle attribute model as a constraint condition to train the geological data pseudo-3D conversion model, so as to obtain the trained geological data pseudo-3D conversion model.

[0040] Step 61: The encoder extracts features layer by layer from the low-frequency pseudo-3D data through 8 convolutional layers and outputs a feature image. Step 62: The decoder uses the dip angle attribute model as a constraint and upsamples the feature image layer by layer through transposed convolution. After the coal seam is upsampled, it is concatenated with the corresponding layer features of the encoder through a skip connection layer to output high-frequency pseudo-3D data. Step 63: Input the three-dimensional amplitude data volume and the high-frequency pseudo-three-dimensional data into the discriminator; The discriminator uses convolutional layers and a binary cross-entropy loss function to compress the feature dimension to 1×1×128 for classification; In this step, the discriminator's task is to distinguish whether the input data is "real" or "generated." The discriminator structure described in the patent employs the following steps: Convolutional layers: Spatial features of input data (high-frequency pseudo-3D data or low-frequency data) are extracted step by step through multiple convolutional layers, while the size (height and width) of the feature map is gradually reduced through stride or pooling.

[0041] Compression to 1×1×128: ​​After a series of convolution operations, the spatial dimension of the feature map is compressed to 1×1 (i.e., one pixel), but 128 channels are retained. This is equivalent to mapping the input data into a feature vector of length 128, condensing global information.

[0042] Classification: This 1×1×128 feature vector is then passed through a fully connected layer or a 1×1 convolutional layer to further map it into a scalar (probability between 0 and 1), representing the likelihood that the input is real data.

[0043] Binary cross-entropy loss: During training, the discriminator's output is compared with the true labels (true data is 1, generated data is 0) to calculate the binary cross-entropy loss.

[0044] In subsequent steps, by minimizing this loss, the discriminator learns to accurately distinguish between true and false data; meanwhile, the generator attempts to generate data that can fool the discriminator, thus forming adversarial training.

[0045] Step 64: Establish a total loss function that includes data distribution loss, structural consistency loss, and spectrum matching loss. Calculate the total loss value and the loss function. As shown in the following formula:

[0046] in, Represents the data distribution loss function; Represents the structural consistency loss function; Represents the spectrum matching loss function; Step 65: With the goal of minimizing the total loss function value, train the pseudo-3D transformation model of geological data until the model converges, and obtain the trained pseudo-3D transformation model of geological data.

[0047] Specifically, in step 63, the data distribution loss function is as follows:

[0048] in, Represents the mathematical expectation operator; This represents the probability that the discriminator will distinguish the input data; specifically, The output is a real number; Represents high-frequency pseudo-3D data; This represents the low-frequency pseudo-3D data output by the decoder; This represents low-frequency pseudo-3D data; The structural consistency loss function is as follows:

[0049] in, Represents the spatial gradient operator for the tilt field; This represents the high-frequency pseudo-3D data output by the generator; Represents the L1 norm; The spectrum matching loss function is as follows:

[0050] in, Indicates the weighting coefficient; Indicates Fourier transform; Represents the L2 norm; Step 5: Construct a pseudo-3D conversion model of geological data, including an encoder, decoder, and discriminator; The encoder includes eight convolutional layers; The decoder is a deconvolution structure symmetrical to the encoder; The discriminator described above employs a multi-layer convolutional structure and a binary cross-entropy loss function. Step 6: Use low-frequency pseudo-3D data as input to the geological data pseudo-3D conversion model, dip angle attribute model as constraint condition, and high-frequency pseudo-3D data as output to the geological data pseudo-3D conversion model; train the geological data pseudo-3D conversion model to obtain the trained geological data pseudo-3D conversion model.

[0051] A pseudo-3D conversion system for geological data includes a data acquisition and preprocessing module, a geological geometric modeling module, a spatial constraint correction module, a differential variable calculation module, a deep learning module, and a dynamic fusion output module; The data acquisition and preprocessing module is used for: Collect geological data from coal mine working faces; Delineate a three-dimensional geometric framework that can cover the coal mine working face; Amplitude balancing processing is performed on the three-dimensional amplitude data volume; The geological geometry modeling module is used for: The three-dimensional geometric framework is divided into multiple three-dimensional meshes of the same size to form an initial geometric model; Calculate the normal vector of the sampling point, obtain the dip angle based on the normal vector, fill the dip angle value into the corresponding sampling point to obtain the dip angle attribute volume; average the dip angle attribute volume to the initial geometric model to form a three-dimensional dip angle attribute model covering the coal mine working face; The spatial constraint correction module is used for: Based on the waveform of the marker layer, trace the marker layer surface in the initial geometric model; Using the marker layer surface as an index, extract the marker layer dip angle attribute value from the three-dimensional dip angle attribute model; The horizontal projection correction of the two-dimensional excavation data is performed based on the dip angle attribute value of the marker layer to obtain the horizontal component of the two-dimensional excavation data. The variational interpolation calculation module is used for: Configure the interpolation objective function with minimizing the value of the interpolation objective function as the baseline condition, and perform three-dimensional interpolation for each three-dimensional grid to obtain low-frequency pseudo-three-dimensional data; The deep learning generation module has a built-in trained geological data pseudo-3D conversion model, which generates high-frequency pseudo-3D data based on low-frequency pseudo-3D data and a 3D dip property model. The dynamic fusion output module is used to fuse low-frequency pseudo-3D data and high-frequency pseudo-3D data to output pseudo-3D geological data volume.

[0052] In this step, the dynamic fusion output module is merged. , The anisotropic filtering method is used to filter out the noise generated during the processing and output pseudo-3D geological data volume.

Claims

1. A method for constructing a pseudo-3D conversion model of geological data, characterized in that, Includes the following steps: Step 1: For the coal mine working face to be processed, delineate the three-dimensional geometric framework that can cover the coal mine working face; Based on the geological data of the coal mine working face, three-dimensional amplitude data volume, two-dimensional excavation data and waveform of marker strata were obtained respectively; The three-dimensional amplitude data volume includes multiple seismic traces distributed in three-dimensional space, with multiple sampling points distributed on each seismic trace, and each sampling point corresponding to an amplitude value. Step 2: Divide the three-dimensional geometric framework into multiple three-dimensional meshes of the same size to form the initial geometric model; For the three-dimensional amplitude data volume, amplitude balancing is performed. For any sampling point in the amplitude-balanced three-dimensional amplitude data volume, its normal vector is calculated. The tilt angle is obtained based on the normal vector, and then the tilt angle value is filled into the corresponding sampling point to obtain the tilt angle attribute volume. Traverse all sampling points and average the dip angle attribute volume to the initial geometric model to form a three-dimensional dip angle attribute model covering the coal mine working face; Step 3: Based on the waveform of the marker layer, trace the marker layer surface in the initial geometric model; Next, using the marker layer surface as an index, the dip angle attribute values ​​of the marker layer are extracted from the three-dimensional dip angle attribute model; The horizontal projection correction of the two-dimensional excavation data is performed based on the dip angle attribute value of the marker layer to obtain the horizontal component of the two-dimensional excavation data; Step 4: Based on the three-dimensional vector of the seismic trace of the three-dimensional amplitude data volume after amplitude balance and the horizontal component of the two-dimensional excavation data, establish the interpolation objective function to calculate the pseudo-three-dimensional data, as shown in the following formula; in, This represents the pseudo-3D data to be solved; Represents the spatial gradient operator for the tilt field; Represents a double integral; This represents the data fitting weights, ranging from 0.01 to 10. Represents the horizontal component of two-dimensional in-situ data; A three-dimensional seismic vector representing the seismic data after amplitude equilibrium; Indicates the index of the sampling point; The dip field spatial gradient operator is represented as follows: in, Indicates the partial derivative sign; It represents the horizontal coordinates, specifically the horizontal coordinates along the horizontal axis within the coal mine working face; This represents the horizontal coordinates, specifically the inner edge of the coal mine working face, dip direction, and direction. Horizontal coordinates on orthogonal horizontal axes; Indicates the dip angle of the strata; Using the minimization of the interpolation objective function as the baseline condition, three-dimensional interpolation is performed for each three-dimensional grid to obtain low-frequency pseudo-three-dimensional data. Step 5: Construct a pseudo-3D conversion model of geological data, including an encoder, decoder, and discriminator; The encoder includes multiple convolutional layers; The decoder is a transposed convolutional structure symmetric to the encoder; The discriminator described above employs a multi-layer convolutional structure and a binary cross-entropy loss function. Step 6: Use low-frequency pseudo-3D data as input to the geological data pseudo-3D conversion model and dip angle attribute model as constraint condition to train the geological data pseudo-3D conversion model and obtain the trained geological data pseudo-3D conversion model.

2. The method for constructing a pseudo-3D conversion model of geological data as described in claim 1, characterized in that, Step 6 specifically includes the following sub-steps: Step 61: The encoder extracts features from the low-frequency pseudo-3D data layer by layer through multiple convolutional layers and outputs a feature image. Step 62: The decoder uses the tilt attribute model as a constraint and upsamples the feature image layer by layer through transposed convolution. After upsampling, it splices the data through skip connection layers to output high-frequency pseudo-3D data. Step 63: Input the three-dimensional amplitude data volume and high-frequency pseudo-three-dimensional data into the discriminator. The discriminator uses a convolutional layer and a binary cross-entropy loss function to compress the feature dimension to 1×1×128 for classification. Step 64: Establish a total loss function that includes data distribution loss, structural consistency loss, and spectrum matching loss. Calculate the total loss value and the loss function. As shown in the following formula: in, Represents the data distribution loss function; Represents the structural consistency loss function; Represents the spectrum matching loss function; Step 65: With the goal of minimizing the total loss function value, train the pseudo-3D transformation model of geological data until the model converges, and obtain the trained pseudo-3D transformation model of geological data.

3. The pseudo-3D conversion method for geological data of coal mine working faces as described in claim 2, characterized in that, In step 63, the data distribution loss function is as follows: in, Represents the mathematical expectation operator; This represents the probability that the discriminator will distinguish the input data; Represents a three-dimensional amplitude data volume; This indicates that the decoder takes low-frequency pseudo-3D data as input and outputs high-frequency pseudo-3D data. This represents low-frequency pseudo-3D data; The structural consistency loss function is as follows: in, Represents the spatial gradient operator for the tilt field; This represents the high-frequency pseudo-3D data output by the decoder, i.e. ; Represents the L1 norm; The spectrum matching loss function is as follows: in, Indicates the weighting coefficient; Indicates Fourier transform; This represents the L2 norm.

4. The pseudo-3D conversion method for geological data of coal mine working faces as described in claim 3, characterized in that, Step 2 includes the following sub-steps: Step 21: Divide the three-dimensional geometric framework into multiple three-dimensional meshes of the same size to form an initial geometric model; Step 22: Apply a normalization method to the amplitude values ​​of each seismic trace in the 3D amplitude data volume to perform amplitude balancing, obtaining the 3D amplitude data volume after amplitude balancing; specifically, the normalization coefficient is as follows: in, Represents the normalization coefficient; Indicates the longitudinal position of the sampling point; Indicates the amplitude of the sampling points on the seismic trace; Indicates the root mean square amplitude; Specifically, in, Indicates the length of the sliding window; Indicates the current sampling point; Step 23: For any sampling point in the three-dimensional amplitude data volume after amplitude balance, calculate its normal vector using the gradient method, determine the tilt angle based on the normal vector, fill the tilt angle value into the corresponding sampling point, and obtain the tilt angle attribute volume; Step 24: Repeat step 23 until every sampling point has been traversed; average the dip angle attribute volume onto the initial geometric model to form a three-dimensional dip angle attribute model covering the coal mine working face. .

5. A pseudo-3D conversion system for geological data, characterized in that, It includes a data acquisition and preprocessing module, a geological geometric modeling module, a spatial constraint correction module, a differential variable calculation module, a deep learning module, and a dynamic fusion output module; The data acquisition and preprocessing module is used for: Collect geological data from coal mine working faces; Delineate a three-dimensional geometric framework that can cover the coal mine working face; Amplitude balancing processing is performed on the three-dimensional amplitude data volume; The geological geometric modeling module is used for: The three-dimensional geometric framework is divided into multiple three-dimensional meshes of the same size to form an initial geometric model; Calculate the normal vector of the sampling point, obtain the dip angle based on the normal vector, fill the dip angle value into the corresponding sampling point to obtain the dip angle attribute volume; average the dip angle attribute volume to the initial geometric model to form a three-dimensional dip angle attribute model covering the coal mine working face; The aforementioned spatial constraint correction module is used for: Based on the waveform of the marker layer, trace the marker layer surface in the initial geometric model; Using the marker layer surface as an index, extract the marker layer dip angle attribute value from the three-dimensional dip angle attribute model; The horizontal projection correction of the two-dimensional excavation data is performed based on the dip angle attribute value of the marker layer to obtain the horizontal component of the two-dimensional excavation data. The variational interpolation calculation module is used for: Configure the interpolation objective function with minimizing the value of the interpolation objective function as the baseline condition, and perform three-dimensional interpolation for each three-dimensional grid to obtain low-frequency pseudo-three-dimensional data; The deep learning generation module has a built-in trained geological data pseudo-3D conversion model, which generates high-frequency pseudo-3D data based on low-frequency pseudo-3D data and a 3D dip angle attribute model. The dynamic fusion output module is used to fuse low-frequency pseudo-3D data and high-frequency pseudo-3D data to output a pseudo-3D geological data volume.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 4.

Images