Method and system for predicting underground physical characteristics by combining seismic waves and drilling data

Abstract

The invention discloses an underground physical characteristic prediction method and system combining seismic waves and drilling data, and relates to the technical field of geophysical exploration, and the method comprises the steps: collecting original seismic records, carrying out the preprocessing, obtaining a three-dimensional seismic data body, carrying out the complex channel analysis of the three-dimensional seismic data body, extracting the instantaneous amplitude, instantaneous phase and instantaneous frequency attributes, and carrying out the calculation of the instantaneous amplitude, instantaneous phase and instantaneous frequency attributes. Fault features are identified through a coherent body algorithm, an earthquake feature matrix is obtained, depth alignment and correction are carried out on density logging, interval transit time and neutron porosity curves of drill holes, and a standardized drill hole data table is generated; and based on the seismic feature matrix and the standardized drilling data table. According to the method, deep complementation of the earthquake macroscopic information and the borehole local truth value is realized through the geometric perception multi-modal graph neural network and the dynamic gating fusion mechanism, the problems of insufficient cross-modal feature fusion and misalignment of physical rules in the prior art are solved, and the prediction precision and reliability under complex geological conditions are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration technology, and in particular to a method and system for predicting underground physical characteristics by combining seismic waves and borehole data. Background Technology

[0002] Accurate prediction of subsurface physical properties is a core aspect of oil and gas exploration, mineral resource assessment, and geological engineering construction. Traditional methods primarily rely on inversion interpretation using single seismic wave data or borehole data. Among these, seismic wave velocity inversion techniques are widely used due to their broad coverage and strong spatial continuity. For example, parameters such as wave impedance and P-wave / S-wave velocity ratio can be obtained through pre-stack and post-stack inversions, thereby inferring lithology and fluid distribution. However, seismic data is limited by the ambiguity of wave velocity-lithology relationships, especially in complex tectonic zones, where single seismic data struggles to achieve precise quantitative analysis of physical properties. Meanwhile, while borehole data can provide high-precision true values, its spatial distribution is sparse due to drilling costs, and its lateral continuity relies on interpolation, easily introducing extrapolation errors.

[0003] The main shortcoming of existing technologies lies in the insufficient depth of cross-modal feature fusion. Current methods typically simply concatenate seismic attributes and borehole parameters before inputting them into traditional machine learning models, ignoring the dynamic interaction mechanism between the spatial geometric features of seismic data and the vertical high-resolution characteristics of borehole data. When constructing the graph structure, existing technologies often use fixed thresholds to define node connection relationships, making it difficult to adaptively adjust the contribution ratio between macroscopic seismic information and local borehole ground truth, resulting in a sharp drop in prediction accuracy outside the well control zone. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for predicting underground physical characteristics by combining seismic waves and borehole data, which solves the problems of insufficient cross-modal feature fusion depth and inability to adaptively balance macroscopic seismic information and local borehole truth in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for predicting underground physical characteristics by combining seismic waves and borehole data, which includes: acquiring raw seismic records and preprocessing them to obtain a three-dimensional seismic data volume; performing complex channel analysis on the three-dimensional seismic data volume to extract instantaneous amplitude, instantaneous phase and instantaneous frequency attributes; identifying fault features through a coherence volume algorithm to obtain a seismic feature matrix; and performing depth alignment and correction on the density logging, sonic transit time and neutron porosity curves of the borehole to generate a standardized borehole data table. Based on the seismic feature matrix and standardized borehole data table, the distance and angle parameters between nodes are calculated, the curvature features are obtained through the eigenvalues ​​of the Hessian matrix, and the dip angle field and azimuth field of the strata are generated by spatial interpolation to obtain the feature tensor of the geometric label. By establishing connecting edges between seismic data points and borehole data points in the feature tensor of the geometric labels based on spatial distance and feature similarity, a graph structure of seismic nodes, borehole nodes, and three types of edges is obtained. A dual-channel graph attention network is used for feature fusion on the graph structures of seismic nodes, borehole nodes, and three types of edges. The feature update direction is dynamically adjusted by embedding a rock physics constraint module to obtain the fused feature tensor. The fused feature tensor is input into the prediction network to generate density, porosity and lithology probability predictions, and the moving cube algorithm is used to construct a three-dimensional geological model. Error calculations were performed on the three-dimensional geological model using data from reserved verification wells, resulting in a comprehensive report.

[0007] As a preferred embodiment of the subsurface physical characteristic prediction method combining seismic waves and borehole data described in this invention, the method includes: acquiring raw seismic records and preprocessing them to obtain a three-dimensional seismic data volume; performing complex-channel analysis on the three-dimensional seismic data volume to extract instantaneous amplitude, instantaneous phase, and instantaneous frequency attributes; identifying fault characteristics using a coherence volume algorithm to obtain a seismic feature matrix; and performing depth alignment and correction on the borehole density logging, sonic transit time, and neutron porosity curves to generate a standardized borehole data table. The method comprises the following steps: Seismic waves are generated by a controllable source, and the reflected signals are received by a detector array to obtain the original seismic record. Adaptive median filtering is applied to the original seismic record to eliminate random noise and obtain denoised seismic data. Refraction static correction is applied to the denoised seismic data to eliminate the influence of surface elevation and obtain the preliminarily corrected seismic gathers. Dynamic correction is then performed based on the stacking velocity scan to eliminate the influence of offset and obtain the NMO-corrected seismic gathers. The NMO-corrected seismic gathers are weighted and stacked to enhance the signal-to-noise ratio, resulting in a three-dimensional seismic data volume. The three-dimensional seismic data volume is then subjected to Hilbert transform to extract instantaneous amplitude, instantaneous phase, and instantaneous frequency attributes, resulting in an instantaneous attribute data volume. Based on the instantaneous attribute data volume, the C3 coherence coefficient is calculated to identify fault characteristics, resulting in a seismic feature matrix with fault enhancement. The dynamic time warping method is used to match the logging curves with the seismic time-depth relationship to obtain depth-aligned logging data. Density compensation and sonic transit time correction are then performed on the depth-aligned logging data to obtain a standardized borehole data table.

[0008] As a preferred embodiment of the subsurface physical characteristic prediction method combining seismic waves and borehole data described in this invention, the method includes the following steps: calculating the distance and angle parameters between nodes based on the seismic feature matrix and standardized borehole data table; obtaining curvature features through Hessian matrix eigenvalues; generating a formation dip field and azimuth field using spatial interpolation; and obtaining a feature tensor with geometric labels. Based on the spatial coordinates in the seismic feature matrix, the Euclidean distance between all seismic data points is calculated to obtain the seismic node distance matrix. Based on the coordinate data in the standardized borehole data table, the horizontal distance between all borehole points is calculated to obtain the borehole node distance matrix. The spatial angle between adjacent nodes is calculated using the coordinates of the seismic nodes, resulting in an angle matrix; Based on the velocity field in the seismic feature matrix, the second-order spatial derivative of each node is obtained, and the Hessian matrix is ​​derived. Eigenvalue decomposition is performed on the Hessian matrix, and the maximum curvature eigenvalue is calculated to obtain the curvature field. By combining curvature field and seismic node elevation data, the local dip angle of the strata is calculated to obtain the dip field; The dip direction of the strata is derived from the dip field, and the azimuth field is output. The Kriging algorithm is used to interpolate the sparse borehole node geometric features to the seismic grid, and the complete geometric field is output. The seismic node distance matrix, borehole node distance matrix, curvature field, dip field and azimuth field are concatenated along the feature dimension to obtain the feature tensor of the geometric label.

[0009] As a preferred embodiment of the subsurface physical characteristic prediction method combining seismic waves and borehole data described in this invention, the method involves: establishing connecting edges between seismic data points and borehole data points in the feature tensor of geometric labels based on spatial distance and feature similarity, thereby obtaining a graph structure of seismic nodes, borehole nodes, and three types of edges, including the following steps. Z-score normalization is performed on the seismic node features in the geometric label feature tensor to output a normalized seismic node feature matrix. Min-Max normalization is performed on the borehole node features in the normalized borehole data table to output a normalized borehole node feature matrix. Based on the normalized seismic node coordinates, spatial neighborhood relationships are established and a seismic node adjacency list is output. Connect adjacent logging points along the borehole depth direction to output the set of borehole longitudinal edges. Find the three nearest neighbor seismic nodes for each borehole node to obtain cross-modal connection pairs. Calculate the edge weights between seismic nodes based on spatial distance and feature similarity. Seismic-borehole edge weights are calculated by combining distance and feature differences; By integrating the adjacency list of seismic nodes, the set of borehole longitudinal edges, and the cross-modal connection pairs, a graph structure of seismic nodes, borehole nodes, and three types of edges is obtained.

[0010] As a preferred embodiment of the subsurface physical characteristic prediction method combining seismic waves and borehole data described in this invention, the method employs a dual-channel graph attention network for feature fusion on the graph structure of seismic nodes, borehole nodes, and three types of edges. The feature update direction is dynamically adjusted by embedding a rock physics constraint module to obtain the fused feature tensor. This includes the following steps: Z-score normalization and Min-Max normalization are applied to the features of seismic nodes and borehole nodes, respectively, to output a node feature matrix with uniform dimensions. The attention weights between seismic nodes are calculated using the GATv2 network to output the higher-order features of the seismic nodes. By aggregating borehole node neighborhood information through the Edge-GAT network, high-order features of borehole nodes are obtained, cross-modal attention weights between seismic and borehole nodes are calculated, and preliminary fusion features are output. By dynamically balancing the contributions of seismic and borehole features through a gating mechanism, the gating fusion features are output, and the density-velocity relationship loss is calculated based on the Gardner equation to obtain the physical error gradient. The fused features are corrected using physical constraint gradients, and node-level, well-level, and region-level feature aggregation is performed to obtain the fused feature tensor.

[0011] As a preferred embodiment of the subsurface physical property prediction method combining seismic waves and borehole data described in this invention, the method includes the following steps: inputting a fused feature tensor into a prediction network to generate density, porosity, and lithology probability prediction values; and constructing a three-dimensional geological model using a moving cubes algorithm. Batch normalization is performed on the fused feature tensor to output standardized features, and the density value of each node is calculated through a fully connected layer. Porosity is calculated using a sigmoid-activated neural network layer to obtain the porosity field; The lithology probability distribution is calculated using a Softmax classifier, and the lithology probability matrix is ​​output. The lithology probability matrix is ​​interpolated onto a regular mesh to obtain 3D volume data. The moving cube algorithm is applied to extract isosurfaces from the volume data to obtain a preliminary triangular mesh model. The mesh topology is optimized using the Laplacian smoothing and edge folding algorithms to output a simplified triangular patch model. The porosity and lithology data are baked into texture maps to obtain a 3D geological model.

[0012] As a preferred embodiment of the subsurface physical characteristic prediction method combining seismic waves and borehole data described in this invention, the method includes the following steps: calculating the error of a three-dimensional geological model using data from reserved verification wells to obtain a comprehensive report. Align the reserved borehole data with the 3D model space, and establish a mapping relationship through time-depth transformation and nearest neighbor matching to obtain the registered verification dataset; The error gradient distribution is calculated along the well trajectory and formation plane, and an error heat map is output. The 3σ criterion is used to identify the error exceeding the limit area, and a list of coordinates and deviation values ​​of the abnormal area are output. Based on the error distribution, retraining and local correction schemes are proposed, and correction recommendations are obtained. The correction recommendations are integrated to obtain a comprehensive report.

[0013] Secondly, the present invention provides a subsurface physical characteristic prediction system combining seismic waves and borehole data, including a data standardization module, which collects raw seismic records and preprocesses them to obtain a three-dimensional seismic data volume, performs complex channel analysis on the three-dimensional seismic data volume to extract instantaneous amplitude, instantaneous phase and instantaneous frequency attributes, identifies fault features through coherence volume algorithm to obtain seismic feature matrix, and performs depth alignment and correction on the borehole density logging, sonic transit time and neutron porosity curves to generate a standardized borehole data table; The feature acquisition module calculates the distance and angle parameters between nodes based on the seismic feature matrix and the standardized borehole data table. It obtains the curvature features through the eigenvalues ​​of the Hessian matrix and generates the dip field and azimuth field of the strata by combining spatial interpolation to obtain the feature tensor of the geometric label. The edge connection module establishes edges between seismic data points and borehole data points in the feature tensor of geometric labels based on spatial distance and feature similarity, resulting in a graph structure of seismic nodes, borehole nodes, and three types of edges. The feature tensor fusion module uses a dual-channel graph attention network to fuse features in the graph structures of seismic nodes, borehole nodes, and three types of edges. The feature update direction is dynamically adjusted by embedding a rock physics constraint module to obtain the fused feature tensor. The 3D geological model module integrates feature tensors into the prediction network to generate density, porosity, and lithology probability predictions, and uses the moving cube algorithm to construct the 3D geological model. The comprehensive report module uses reserved verification well data to calculate errors in the three-dimensional geological model and obtain a comprehensive report.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the method for predicting underground physical characteristics by combining seismic waves and borehole data as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for predicting underground physical characteristics by combining seismic waves and borehole data as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: A three-dimensional seismic data volume is obtained by acquiring and preprocessing seismic data; instantaneous attributes are extracted and fault features are identified to generate a seismic feature matrix; borehole logging data is depth-aligned and corrected to generate a standardized data table; node distance, angle, and curvature features are calculated based on the seismic feature matrix and borehole data to generate a geometric label feature tensor; a graph structure containing seismic nodes, borehole nodes, and three types of edges is constructed; a dual-channel graph attention network is used for feature fusion and rock physics constraints are embedded to obtain a fused feature tensor; a three-dimensional geological model is constructed by generating density, porosity, and lithology probabilities through a prediction network; finally, error calculation and model correction are performed using validation well data. Through a geometrically perceptive multimodal graph neural network and a dynamic gating fusion mechanism, deep complementarity between macroscopic seismic information and local borehole ground truth is achieved, solving the problems of insufficient cross-modal feature fusion and inaccurate physical laws in existing technologies, and significantly improving the prediction accuracy and reliability under complex geological conditions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for predicting underground physical characteristics by combining seismic waves and borehole data.

[0019] Figure 2 This is a schematic diagram of a system for predicting underground physical characteristics by combining seismic waves and borehole data. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Reference Figures 1-2 This is one embodiment of the present invention, which provides a method for predicting subsurface physical characteristics by combining seismic waves and borehole data, including the following steps: S1. Acquire raw seismic records and preprocess them to obtain a three-dimensional seismic data volume. Perform complex channel analysis on the three-dimensional seismic data volume to extract instantaneous amplitude, instantaneous phase, and instantaneous frequency attributes. Identify fault features through the coherence volume algorithm to obtain the seismic feature matrix. Perform depth alignment and correction on the borehole density logging, sonic transit time, and neutron porosity curves to generate a standardized borehole data table.

[0024] S1.1. Seismic waves are generated by a controllable seismic source, and the reflected signals are received by a detector array to obtain the original seismic record. Adaptive median filtering is performed on the original seismic record to eliminate random noise and obtain denoised seismic data.

[0025] Furthermore, broadband seismic waves (dominant frequency range 10-80Hz) are excited by a controllable source (such as a hydraulic vibrator or an explosive source). A regular grid array of geophones (spaced 25-50m) is used to receive the reflected signals, and raw seismic records (SEGY format) are acquired. The raw records are then subjected to adaptive median filtering, and the filtering intensity is dynamically adjusted according to the signal-to-noise ratio of the signal within the sliding window. This effectively suppresses random noise while retaining the high-frequency components of the effective reflected waves, and outputs seismic data with improved signal-to-noise ratio.

[0026] S1.2 Apply static refraction correction to the denoised seismic data to eliminate the influence of surface elevation and obtain a preliminarily corrected seismic gather. Perform dynamic correction based on stacking velocity scan to eliminate the influence of offset and obtain an NMO-corrected seismic gather.

[0027] Furthermore, static refraction correction is applied to the denoised seismic data to eliminate time differences caused by topographic undulations and variations in regolith thickness. Subsequently, velocity analysis and dynamic correction are performed. The optimal stacking velocity field is determined through iterative velocity scanning, and the hyperbolic time-distance equation is applied to correct time differences in non-zero offset gathers, eliminating the influence of shot-receiver offset on reflected wave travel time. The resulting seismic gathers, processed by both static and dynamic corrections, are then output.

[0028] S1.3. Weighted stacking of NMO-corrected seismic gathers is performed to enhance the signal-to-noise ratio, resulting in a three-dimensional seismic data volume. Hilbert transform is applied to the three-dimensional seismic data volume to extract instantaneous amplitude, instantaneous phase, and instantaneous frequency attributes, resulting in an instantaneous attribute data volume. Based on the instantaneous attribute data volume, the C3 coherence coefficient is calculated to identify fault characteristics, resulting in a fault-enhanced seismic feature matrix.

[0029] Specifically, the expression is, ; in, As a measure of coherence, For earthquake data in the main direction, This represents the intensity of changes in earthquake data in secondary directions. This represents the variation of earthquake data in the weakest direction.

[0030] S1.4. The dynamic time warping method is used to match the logging curve with the seismic time-depth relationship to obtain depth-aligned logging data. Density compensation and sonic transit time correction are performed on the depth-aligned logging data to obtain a standardized borehole data table.

[0031] Furthermore, a dynamic time warping algorithm is employed to perform non-rigid time-depth matching between well logging curves and well-side seismic traces. Dynamic programming is used to find the optimal path, eliminating time-depth conversion deviations caused by velocity model errors and ensuring precise alignment between well logging data and the seismic time domain. Subsequently, environmental corrections are performed on the depth-aligned well logging data: for density curves, a wellbore compensation model is applied to eliminate the influence of wellbore enlargement; for sonic transit time curves, a mud invasion model is used to correct for the flushing zone effect, and the correction results are verified using laboratory core data. Finally, a standardized borehole data table is output.

[0032] S2. Based on the seismic feature matrix and standardized borehole data table, calculate the distance and angle parameters between nodes, obtain the curvature features through the eigenvalues ​​of the Hessian matrix, and generate the dip field and azimuth field of the strata by combining spatial interpolation to obtain the feature tensor of the geometric label.

[0033] S2.1. Based on the spatial coordinates in the seismic feature matrix, calculate the Euclidean distance between all seismic data points to obtain the seismic node distance matrix.

[0034] Specifically, the expression is, ; in, Current earthquake node With the target node The three-dimensional Euclidean distance between them Current earthquake node Eastward coordinates Current earthquake node North coordinates Current earthquake node Vertical depth For target earthquake nodes Eastward coordinates For target earthquake nodes North coordinates For target earthquake nodes Vertical depth.

[0035] S2.2 Based on the coordinate data in the standardized borehole data table, calculate the horizontal distance between all borehole points to obtain the borehole node distance matrix.

[0036] Specifically, the expression is, ; in, The current node of the borehole To the target node The horizontal distance between them The current node of the borehole Eastward coordinates For drilling target node Eastward coordinates The current node of the borehole North coordinates The current node of the borehole North-facing coordinates.

[0037] S2.3 Calculate the spatial angle between adjacent nodes using the coordinates of the seismic nodes to obtain the angle matrix.

[0038] Specifically, the expression is, ; in, For the current node and target node The spatial angle, For the current node spatial vectors, For the target node . spatial vector.

[0039] S2.4. Based on the velocity field in the seismic feature matrix, the second-order spatial derivative of each node is obtained, and the Hessian matrix is ​​obtained.

[0040] Furthermore, based on the P-wave velocity field in the seismic feature matrix, a Gaussian second-order derivative filter is used to calculate the second-order partial derivative of each grid node in three-dimensional space, constructing a Hessian matrix. During the calculation, a scale-adaptive Gaussian kernel is used for smoothing to suppress high-frequency noise interference in the velocity field, while preserving the curvature variation characteristics caused by faults, folds, and other structures. The final output Hessian matrix characterizes the local geometry of the velocity field in space.

[0041] The curvature field is obtained by performing eigenvalue decomposition on the Hessian matrix and calculating the maximum curvature eigenvalue.

[0042] Specifically, the expression is, ; in, The maximum curvature eigenvalue, For the Hessian matrix locus, Let be the determinant of the Hessian matrix.

[0043] S2.5. Combining the curvature field and seismic node elevation data, calculate the local dip angle of the strata to obtain the dip field.

[0044] Specifically, the expression is, ; in, This represents the local dip angle of the strata. For elevation in Rate of change of the axis For elevation in Rate of change of the axis For elevation in Rate of change of the axis.

[0045] S2.6. Based on the dip field, the dip direction of the strata is obtained, and the azimuth field is output. The Kriging algorithm is used to interpolate the sparse borehole node geometric features to the seismic grid and output the complete geometric field. The seismic node distance matrix, borehole node distance matrix, curvature field, dip field and azimuth field are spliced ​​along the feature dimension to obtain the feature tensor of the geometric label.

[0046] Furthermore, based on the dip field, the horizontal projection direction of the stratum dip is calculated, and the azimuth is determined by the four-quadrant arctangent function to eliminate the directional ambiguity of the traditional arctangent function and generate a continuous and stable azimuth field. Then, the Kriging interpolation algorithm is used, with the geometric characteristics of the borehole nodes as regional variables. Based on the spherical model, the spatial variogram is fitted, and the sparse borehole data is interpolated to the seismic grid nodes to ensure that the interpolation results conform to the laws of geostatistics. Finally, the distance matrix, curvature field, dip field and azimuth field of the seismic and borehole data are stitched together according to the feature dimensions.

[0047] S3. Connect the seismic data points and borehole data points in the feature tensor of the geometric labels according to spatial distance and feature similarity to obtain the graph structure of seismic nodes, borehole nodes and three types of edges.

[0048] S3.1 Perform Z-score normalization on the seismic node features in the geometric label feature tensor and output the normalized seismic node feature matrix. Perform Min-Max normalization on the borehole node features in the normalized borehole data table and output the normalized borehole node feature matrix. Based on the normalized seismic node coordinates, establish spatial neighborhood relationships and output the seismic node adjacency list.

[0049] Furthermore, Z-score standardization is applied to the seismic node features in the geometric label feature tensor. By calculating the mean and standard deviation of each feature dimension, the influence of different physical dimensions is eliminated, ensuring that the numerical distribution of the seismic node feature matrix conforms to a standard normal distribution with zero mean and unit variance. Simultaneously, Min-Max normalization is performed on the borehole node features in the standardized borehole data table, linearly mapping each logging parameter to the [0,1] interval, preserving the relative relationships of the original data and improving the model training stability. Based on the standardized three-dimensional coordinates of the seismic nodes, a KD-Tree spatial index is used to quickly establish neighborhood connectivity within a 50-meter radius, generating a seismic node adjacency table representing the spatial topology of the nodes.

[0050] S3.2 Connect adjacent logging points along the borehole depth direction, output the set of borehole longitudinal edges, find the three nearest neighbor seismic nodes for each borehole node, obtain cross-modal connection pairs, and calculate the edge weights between seismic nodes based on spatial distance and feature similarity.

[0051] Specifically, the expression is, ; in, Current earthquake node With the target node Edge weights between them Current earthquake node cosine similarity, For target earthquake nodes The cosine similarity.

[0052] S3.3 Calculate the seismic-borehole edge weight by combining distance and feature differences.

[0053] Specifically, the expression is, ; in, For seismic-borehole edge weights.

[0054] S3.4 Integrate the seismic node adjacency list, borehole longitudinal edge set, and cross-modal connection pairs to obtain the graph structure of seismic nodes, borehole nodes, and three types of edges.

[0055] Furthermore, the three types of topological relationships—seismic node adjacency lists, borehole longitudinal edge sets, and cross-modal connections—are integrated to construct a unified multimodal graph structure. Seismic-seismic edges preserve local structural continuity, borehole-borehole edges strengthen vertical high-resolution constraints, and seismic-borehole edges enable cross-modal information transfer. The resulting graph structure comprehensively characterizes the spatial grid association of seismic data, the longitudinal continuity of borehole data, and the complementary mapping relationships between the two types of data.

[0056] S4. A dual-channel graph attention network is used for feature fusion on the graph structure of seismic nodes, borehole nodes and three types of edges. The feature update direction is dynamically adjusted by embedding a rock physics constraint module to obtain the fused feature tensor.

[0057] S4.1 Perform Z-score standardization and Min-Max normalization on the features of seismic nodes and borehole nodes respectively, and output a node feature matrix with uniform dimensions. Use the GATv2 network to calculate the attention weights between seismic nodes and output the high-order features of seismic nodes.

[0058] Furthermore, Z-score standardization is applied to the seismic node features. By subtracting the mean and dividing by the standard deviation, the scale differences of different physical dimensions are eliminated, making the feature distribution meet the requirements of zero mean and unit variance. At the same time, Min-Max normalization is applied to the borehole node features, linearly mapping each logging parameter to the [0,1] interval, preserving the relative relationship of the original data. Based on the standardized node features, the attention weights between seismic nodes are dynamically calculated using a GATv2 graph attention network. The query-key vector is extracted through a multilayer perceptron and the LeakyReLU activation function is applied to generate attention coefficients that reflect the nonlinear correlation between nodes.

[0059] S4.2. Aggregate borehole node neighborhood information through the Edge-GAT network to obtain high-order features of borehole nodes, calculate cross-modal attention weights between seismic and borehole nodes, and output preliminary fusion features.

[0060] Specifically, the expression is, ; in, As a preliminary fusion feature, To query the projection matrix, Current earthquake node Features For drilling nodes Features This is the scaling factor.

[0061] S4.3. Dynamically balance the contributions of seismic and borehole features through a gating mechanism, output gating fusion features, calculate density-velocity relationship loss based on Gardner equation, and obtain physical error gradient.

[0062] Specifically, the expression is, ; in, For the gated vector, For the gated weight matrix, These are high-order features of earthquake nodes. These are high-order features of the borehole nodes.

[0063] S4.4. Use physical constraint gradient correction to refine the fused features, and perform node-level, well-level, and region-level feature aggregation to obtain the fused feature tensor.

[0064] Furthermore, the fused features are directionally corrected by adjusting components in the feature space that violate rock physics laws through backpropagation to ensure that the prediction results conform to the empirical relationship between density and velocity. Subsequently, a three-level feature aggregation strategy is adopted: at the node level, local detailed features are extracted through fully connected layers; at the wellbore level, mean pooling is performed on node features within the same borehole to preserve vertical continuity; and at the regional level, graph pooling is used to aggregate global features within structural units. Finally, a fused feature tensor with both local accuracy and regional consistency is output, providing multi-scale representation for 3D geological modeling.

[0065] S5. Input the fused feature tensor into the prediction network to generate density, porosity and lithology probability prediction values, and use the moving cube algorithm to construct a three-dimensional geological model.

[0066] S5.1 Perform batch normalization on the fused feature tensor to output standardized features, and calculate the density value of each node through a fully connected layer.

[0067] Specifically, the expression is, ; in, The density value of the nodes. This is the weight matrix. The features of the merged nodes, This is a bias term.

[0068] S5.2 Calculate porosity using a sigmoid-activated neural network layer to obtain the porosity field.

[0069] Specifically, the expression is, ; in, To predict porosity.

[0070] S5.3. Use the Softmax classifier to calculate the lithology probability distribution and output the lithology probability matrix.

[0071] Specifically, the expression is, ; in, This represents the lithological probability distribution.

[0072] S5.4. Interpolate the lithology probability matrix to a regular grid to obtain three-dimensional volume data. Apply the moving cube algorithm to extract isosurfaces from the volume data to obtain a preliminary triangular mesh model. Optimize the mesh topology using the Laplacian smoothing and edge folding algorithms to output a simplified triangular patch model. Bake the porosity and lithology data into texture maps to obtain a three-dimensional geological model.

[0073] Furthermore, the lithological probability matrix is ​​mapped onto a regular grid of 25m×25m×2ms using trilinear interpolation to generate three-dimensional volume data containing the probability distributions of sandstone, mudstone, and limestone. Based on the moving cube algorithm, isosurfaces of lithological interfaces are extracted from the three-dimensional volume data to form a preliminary mesh model composed of triangular facets. The preliminary mesh model is then smoothed using Laplacian smoothing, and jagged edges are eliminated by iteratively adjusting vertex positions. An edge-folding algorithm is then used to simplify redundant triangular facets, preserving the main structural features while improving rendering efficiency. Finally, the porosity field and lithological classification results are baked into RGB texture maps to generate a three-dimensional geological model with attribute annotations.

[0074] S6. Calculate the error of the three-dimensional geological model using the data from the reserved verification wells, and obtain a comprehensive report.

[0075] S6.1 Align the reserved borehole data with the 3D model space, establish a mapping relationship through time-depth conversion and nearest neighbor matching, and obtain the registered verification dataset.

[0076] Furthermore, the reserved borehole data is spatially aligned with the 3D geological model, and the well logging depth domain data is converted into the time domain based on the verification velocity model to ensure matching with the seismic time domain model. The model node corresponding to each well logging point is located in the 3D grid through the nearest neighbor search algorithm, and the accurate mapping relationship between well logging data and model nodes is established to obtain the registered verification dataset.

[0077] S6.2 Calculate the error gradient distribution along the well trajectory and formation plane, output the error heat map, use the 3σ criterion to identify the error exceeding the limit area, output the coordinate list of the abnormal area and the deviation value, propose retraining and local correction schemes based on the error distribution, obtain correction suggestions, and integrate the correction suggestions to obtain a comprehensive report.

[0078] Furthermore, the predicted error gradients of parameters such as density and porosity are calculated vertically along the borehole trajectory, and a planar error distribution field is generated by interpolation at the formation level. Color coding is used to generate an error heatmap, which intuitively displays the spatial distribution pattern of errors. Based on the 3σ criterion, areas exceeding the error limit are automatically marked, and the three-dimensional coordinates and deviation of the anomaly area are output. Combining geological understanding, the causes of the anomaly area are classified, and targeted training strategies and local grid correction schemes are proposed. Finally, a comprehensive report containing error analysis, correction suggestions, and effect verification is formed.

[0079] This embodiment also provides a subsurface physical characteristic prediction system that combines seismic waves and borehole data, including: The data standardization module collects raw seismic records and preprocesses them to obtain a three-dimensional seismic data volume. It performs complex-channel analysis on the three-dimensional seismic data volume to extract instantaneous amplitude, instantaneous phase, and instantaneous frequency attributes. It identifies fault features through the coherence volume algorithm to obtain a seismic feature matrix. It also performs depth alignment and correction on the borehole density logging, sonic transit time, and neutron porosity curves to generate a standardized borehole data table. The feature acquisition module calculates the distance and angle parameters between nodes based on the seismic feature matrix and the standardized borehole data table. It obtains the curvature features through the eigenvalues ​​of the Hessian matrix and generates the dip field and azimuth field of the strata by combining spatial interpolation to obtain the feature tensor of the geometric label. The edge connection module establishes edges between seismic data points and borehole data points in the feature tensor of geometric labels based on spatial distance and feature similarity, resulting in a graph structure of seismic nodes, borehole nodes, and three types of edges. The feature tensor fusion module uses a dual-channel graph attention network to fuse features in the graph structures of seismic nodes, borehole nodes, and three types of edges. The feature update direction is dynamically adjusted by embedding a rock physics constraint module to obtain the fused feature tensor. The 3D geological model module integrates feature tensors into the prediction network to generate density, porosity, and lithology probability predictions, and uses the moving cube algorithm to construct the 3D geological model. The comprehensive report module uses reserved verification well data to calculate errors in the three-dimensional geological model and obtain a comprehensive report.

[0080] This embodiment also provides a computer device suitable for the prediction method of underground physical characteristics combining seismic waves and borehole data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the prediction method of underground physical characteristics combining seismic waves and borehole data as proposed in the above embodiment.

[0081] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0082] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for predicting underground physical characteristics by combining seismic waves and borehole data as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0083] In summary, this invention obtains a three-dimensional seismic data volume by acquiring and preprocessing seismic data, extracts instantaneous attributes, and identifies fault features to generate a seismic feature matrix; it performs depth alignment and correction on borehole logging data to generate a standardized data table; based on the seismic feature matrix and borehole data, it calculates node distance, angle, and curvature features to generate a geometric label feature tensor; it constructs a graph structure containing seismic nodes, borehole nodes, and three types of edges; it uses a dual-channel graph attention network for feature fusion and embeds rock physics constraints to obtain a fused feature tensor; it generates density, porosity, and lithology probabilities through a prediction network to construct a three-dimensional geological model; finally, it uses validation well data for error calculation and model correction. Through a geometrically perceptive multimodal graph neural network and a dynamic gating fusion mechanism, it achieves deep complementarity between macroscopic seismic information and local borehole ground truth, solving the problems of insufficient cross-modal feature fusion and inaccurate physical laws in existing technologies, and significantly improving the prediction accuracy and reliability under complex geological conditions.

[0084] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting underground physical characteristics by combining seismic waves and borehole data, characterized in that: include, Raw seismic records are collected and preprocessed to obtain a three-dimensional seismic data volume. Complex trace analysis is performed on the three-dimensional seismic data volume to extract instantaneous amplitude, instantaneous phase, and instantaneous frequency attributes. Fault features are identified through the coherence volume algorithm to obtain the seismic feature matrix. The density logging, sonic transit time, and neutron porosity curves of the boreholes are depth aligned and corrected to generate a standardized borehole data table. Based on the seismic feature matrix and standardized borehole data table, the distance and angle parameters between nodes are calculated, the curvature features are obtained through the eigenvalues ​​of the Hessian matrix, and the dip angle field and azimuth field of the strata are generated by spatial interpolation to obtain the feature tensor of the geometric label. By establishing connecting edges between seismic data points and borehole data points in the feature tensor of the geometric labels based on spatial distance and feature similarity, a graph structure of seismic nodes, borehole nodes, and three types of edges is obtained. A dual-channel graph attention network is used for feature fusion on the graph structures of seismic nodes, borehole nodes, and three types of edges. The feature update direction is dynamically adjusted by embedding a rock physics constraint module to obtain the fused feature tensor. The fused feature tensor is input into the prediction network to generate density, porosity and lithology probability predictions, and the moving cube algorithm is used to construct a three-dimensional geological model. Error calculations were performed on the three-dimensional geological model using data from reserved verification wells, resulting in a comprehensive report.

2. The method for predicting underground physical characteristics by combining seismic waves and borehole data as described in claim 1, characterized in that: Raw seismic records are acquired and preprocessed to obtain a 3D seismic data volume. Complex trace analysis is performed on the 3D seismic data volume to extract instantaneous amplitude, instantaneous phase, and instantaneous frequency attributes. Fault features are identified using a coherence volume algorithm to obtain a seismic feature matrix. Depth alignment and correction are then performed on borehole density logging, sonic transit time, and neutron porosity curves to generate a standardized borehole data table. This process includes the following steps: Seismic waves are generated by a controllable source, and the reflected signals are received by a detector array to obtain the original seismic record. Adaptive median filtering is applied to the original seismic record to eliminate random noise and obtain denoised seismic data. Refraction static correction is applied to the denoised seismic data to eliminate the influence of surface elevation and obtain the preliminarily corrected seismic gathers. Dynamic correction is then performed based on the stacking velocity scan to eliminate the influence of offset and obtain the NMO-corrected seismic gathers. The NMO-corrected seismic gathers are weighted and stacked to enhance the signal-to-noise ratio, resulting in a three-dimensional seismic data volume. The three-dimensional seismic data volume is then subjected to Hilbert transform to extract instantaneous amplitude, instantaneous phase, and instantaneous frequency attributes, resulting in an instantaneous attribute data volume. Based on the instantaneous attribute data volume, the C3 coherence coefficient is calculated to identify fault characteristics and obtain the fault-enhanced seismic characteristic moments. The dynamic time warping method is used to match the logging curves with the seismic time-depth relationship to obtain depth-aligned logging data. Density compensation and sonic transit time correction are then performed on the depth-aligned logging data to obtain a standardized borehole data table.

3. The method for predicting underground physical characteristics by combining seismic waves and borehole data as described in claim 2, characterized in that: Based on the seismic feature matrix and standardized borehole data tables, the distance and angle parameters between nodes are calculated. Curvature features are obtained through the eigenvalues ​​of the Hessian matrix. Combined with spatial interpolation, the dip and azimuth fields of the formation are generated, resulting in the feature tensor with geometric labels. Includes the following steps, Based on the spatial coordinates in the seismic feature matrix, the Euclidean distance between all seismic data points is calculated to obtain the seismic node distance matrix. Based on the coordinate data in the standardized borehole data table, the horizontal distance between all borehole points is calculated to obtain the borehole node distance matrix. The spatial angle between adjacent nodes is calculated using the coordinates of the seismic nodes, resulting in an angle matrix; Based on the velocity field in the seismic feature matrix, the second-order spatial derivative of each node is obtained, and the Hessian matrix is ​​derived. Eigenvalue decomposition is performed on the Hessian matrix, and the maximum curvature eigenvalue is calculated to obtain the curvature field. By combining curvature field and seismic node elevation data, the local dip angle of the strata is calculated to obtain the dip field; The dip direction of the strata is derived from the dip field, and the azimuth field is output. The Kriging algorithm is used to interpolate the sparse borehole node geometric features to the seismic grid, and the complete geometric field is output. The seismic node distance matrix, borehole node distance matrix, curvature field, dip field and azimuth field are concatenated along the feature dimension to obtain the feature tensor of the geometric label.

4. The method for predicting underground physical characteristics by combining seismic waves and borehole data as described in claim 3, characterized in that: The seismic and borehole data points in the feature tensor of the geometric labels are connected by edges based on spatial distance and feature similarity, resulting in a graph structure of seismic nodes, borehole nodes, and three types of edges. This includes the following steps. Z-score normalization is performed on the seismic node features in the geometric label feature tensor to output a normalized seismic node feature matrix. Min-Max normalization is performed on the borehole node features in the normalized borehole data table to output a normalized borehole node feature matrix. Based on the normalized seismic node coordinates, spatial neighborhood relationships are established and a seismic node adjacency list is output. Connect adjacent logging points along the borehole depth direction to output the set of borehole longitudinal edges. Find the three nearest neighbor seismic nodes for each borehole node to obtain cross-modal connection pairs. Calculate the edge weights between seismic nodes based on spatial distance and feature similarity. Seismic-borehole edge weights are calculated by combining distance and feature differences; By integrating the adjacency list of seismic nodes, the set of borehole longitudinal edges, and the cross-modal connection pairs, a graph structure of seismic nodes, borehole nodes, and three types of edges is obtained.

5. The method for predicting underground physical characteristics by combining seismic waves and borehole data as described in claim 4, characterized in that: A dual-channel graph attention network is used for feature fusion on the graph structures of seismic nodes, borehole nodes, and three types of edges. By embedding a rock physics constraint module, the feature update direction is dynamically adjusted to obtain the fused feature tensor. Includes the following steps, Z-score normalization and Min-Max normalization are applied to the features of seismic nodes and borehole nodes, respectively, to output a node feature matrix with uniform dimensions. The attention weights between seismic nodes are calculated using the GATv2 network to output the higher-order features of the seismic nodes. By aggregating borehole node neighborhood information through the Edge-GAT network, high-order features of borehole nodes are obtained, cross-modal attention weights between seismic and borehole nodes are calculated, and preliminary fusion features are output. By dynamically balancing the contributions of seismic and borehole features through a gating mechanism, the gating fusion features are output, and the density-velocity relationship loss is calculated based on the Gardner equation to obtain the physical error gradient. The fused features are corrected using physical constraint gradients, and node-level, well-level, and region-level feature aggregation is performed to obtain the fused feature tensor.

6. The method for predicting underground physical characteristics by combining seismic waves and borehole data as described in claim 5, characterized in that: The fused feature tensor is input into the prediction network to generate density, porosity, and lithology probability predictions. A three-dimensional geological model is then constructed using the moving cubes algorithm, including the following steps. Batch normalization is performed on the fused feature tensor to output standardized features, and the density value of each node is calculated through a fully connected layer. Porosity is calculated using a sigmoid-activated neural network layer to obtain the porosity field; The lithology probability distribution is calculated using a Softmax classifier, and the lithology probability matrix is ​​output. The lithology probability matrix is ​​interpolated onto a regular mesh to obtain 3D volume data. The moving cube algorithm is applied to extract isosurfaces from the volume data to obtain a preliminary triangular mesh model. The mesh topology is optimized using the Laplacian smoothing and edge folding algorithms to output a simplified triangular patch model. The porosity and lithology data are baked into texture maps to obtain a 3D geological model.

7. The method for predicting underground physical characteristics by combining seismic waves and borehole data as described in claim 6, characterized in that: Error calculations were performed on the 3D geological model using data from reserved verification wells to obtain a comprehensive report, including the following steps. Align the reserved borehole data with the 3D model space, and establish a mapping relationship through time-depth transformation and nearest neighbor matching to obtain the registered verification dataset; The error gradient distribution is calculated along the well trajectory and formation plane, and an error heat map is output. The 3σ criterion is used to identify the error exceeding the limit area, and a list of coordinates and deviation values ​​of the abnormal area are output. Based on the error distribution, retraining and local correction schemes are proposed, and correction recommendations are obtained. The correction recommendations are integrated to obtain a comprehensive report.

8. A subsurface physical characteristic prediction system combining seismic waves and borehole data, based on the subsurface physical characteristic prediction method combining seismic waves and borehole data as described in any one of claims 1 to 7, characterized in that: include, The data standardization module collects raw seismic records and preprocesses them to obtain a three-dimensional seismic data volume. It performs complex-channel analysis on the three-dimensional seismic data volume to extract instantaneous amplitude, instantaneous phase, and instantaneous frequency attributes. It identifies fault features through the coherence volume algorithm to obtain a seismic feature matrix. It also performs depth alignment and correction on the borehole density logging, sonic transit time, and neutron porosity curves to generate a standardized borehole data table. The feature acquisition module calculates the distance and angle parameters between nodes based on the seismic feature matrix and the standardized borehole data table. It obtains the curvature features through the eigenvalues ​​of the Hessian matrix and generates the dip field and azimuth field of the strata by combining spatial interpolation to obtain the feature tensor of the geometric label. The edge connection module establishes edges between seismic data points and borehole data points in the feature tensor of geometric labels based on spatial distance and feature similarity, resulting in a graph structure of seismic nodes, borehole nodes, and three types of edges. The feature tensor fusion module uses a dual-channel graph attention network to fuse features in the graph structures of seismic nodes, borehole nodes, and three types of edges. The feature update direction is dynamically adjusted by embedding a rock physics constraint module to obtain the fused feature tensor. The 3D geological model module integrates feature tensors into the prediction network to generate density, porosity, and lithology probability predictions, and uses the moving cube algorithm to construct the 3D geological model. The comprehensive report module uses reserved verification well data to calculate errors in the three-dimensional geological model and obtain a comprehensive report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for predicting underground physical characteristics by combining seismic waves and borehole data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for predicting underground physical characteristics by combining seismic waves and borehole data as described in any one of claims 1 to 7.

Images