A method for separating seismic longitudinal and transverse wave data, a model building method and a system

Abstract

The present invention discloses a method for separating seismic P- and S-wave data, a model construction method, and a system, relating to the technical field of seismic P- and S-wave separation. The model construction method includes constructing several elastic models based on seismic data; the elastic models are determined by a set of medium density, P-wave velocity, S-wave velocity, and anisotropy parameters; the elastic models are processed to determine a sample data set; the sample data set includes multiple sample pairs; the sample pairs include three-component seismic data and corresponding separated P- and S-wave data; and an artificial intelligence network is trained using the sample data set to obtain a seismic P- and S-wave data separation model for separating seismic P- and S-wave data in a survey area. The present invention can achieve the purpose of obtaining high-efficiency, high-resolution P- and S-wave data.

Description

Technical Field

[0001] The present invention relates to the technical field of earthquake P- and S-wave separation, and in particular to a method for separating earthquake P- and S-wave data, a system for separating earthquake P- and S-wave data, a method for constructing an earthquake P- and S-wave data separation model, and a system for constructing an earthquake P- and S-wave data separation model. Background Art

[0002] With the development of multi-wave seismic imaging technology, its imaging processing flow has also changed. Based on the different imaging processing flows, multi-wave seismic imaging technology can be divided into two categories: 1) scalar imaging technology based on wavefield separation technology. This method is based on scalar wavefield theory and first separates multi-component seismic data into longitudinal and shear wave data. These two types of data are then processed separately based on the conventional longitudinal wave processing flow; 2) vector imaging technology combining longitudinal and shear waves. This method is based on vector wavefield theory and processes the collected multi-component seismic data as a vector field. At the same time, wavefield separation technology is used during the implementation of the imaging algorithm to achieve high-precision multi-wave imaging results without crosstalk.

[0003] The two aforementioned multiwave seismic imaging techniques each have their advantages and disadvantages. The first type relies on pure wavefield data, eliminating crosstalk artifacts. Furthermore, due to its scalar wavefield theory, it enjoys high computational efficiency. However, this approach destroys the vectorial nature of the seismic wavefield and ignores the coupling relationship between P- and S-waves, thus hindering the full potential of multiwave seismic imaging. The second type, vector field imaging, which combines P- and S-waves, fully exploits the vectorial nature of the seismic wavefield and fully preserves the coupling relationship between P- and S-waves. Therefore, this method is the inevitable direction for the development of multiwave seismic imaging. In both types of multiwave seismic imaging, P- and S-wave separation is essential. For the first type of imaging, P- and S-wave separation in seismic records is fundamental, and the quality of this separation directly impacts imaging quality. For the second type of imaging, P- and S-wave separation in seismic wavefield snapshots is crucial, and the quality of this separation directly affects the degree of P- and S-wave crosstalk artifacts in the imaging profile. Therefore, research on P- and S-wave separation holds significant practical significance for multiwave seismic imaging.

[0004] Although various methods for separating P- and S-waves from seismic wavefields have been developed, these methods are susceptible to factors such as aliasing, endpoint effects, noise, or analysis time windows. Furthermore, they all rely on accurate medium parameters, resulting in low computational efficiency in three-dimensional problems and significant limitations in practical applications. Therefore, a high-efficiency, high-resolution P- and S-wave separation method that is independent of parameter accuracy is needed. Summary of the Invention

[0005] In view of the above-mentioned defects in the prior art, the present invention provides a seismic P- and S-wave data separation method, a model building method and a system to achieve the purpose of obtaining high-efficiency, high-resolution P- and S-wave data.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] In a first aspect, the present invention provides a method for constructing a seismic P-wave and S-wave data separation model, comprising:

[0008] Collecting seismic data and constructing a plurality of elastic models based on the seismic data; the elastic models are determined by a set of medium density, longitudinal wave velocity, shear wave velocity and anisotropy parameters;

[0009] The elastic model is processed to determine a sample data set; the sample data set includes a plurality of sample pairs; the sample pairs include sample input data and corresponding label data; the sample input data is three-component seismic data; the label data is separated longitudinal and transverse wave data;

[0010] The artificial intelligence network is trained using a sample data set to obtain a seismic P-wave and S-wave data separation model; the seismic P-wave and S-wave data separation model is used to separate the seismic P-wave and S-wave data in the survey area.

[0011] In a second aspect, the present invention provides a system for constructing a seismic P-wave and S-wave data separation model, comprising:

[0012] An elastic model building module, for collecting seismic data and building a plurality of elastic models based on the seismic data; the elastic models are determined by a set of medium density, longitudinal wave velocity, shear wave velocity and anisotropy parameters;

[0013] A sample data set determination module is used to process the elastic model to determine a sample data set; the sample data set includes a plurality of sample pairs; the sample pairs include sample input data and corresponding label data; the sample input data is three-component seismic data; the label data is separated P-wave and S-wave data;

[0014] The earthquake P-wave and S-wave data separation model construction module is used to train the artificial intelligence network using a sample data set to obtain an earthquake P-wave and S-wave data separation model; the earthquake P-wave and S-wave data separation model is used to separate the earthquake P-wave and S-wave data in the survey area.

[0015] In a third aspect, the present invention provides a method for separating seismic P- and S-wave data, comprising:

[0016] Acquire three-component seismic data of the survey area;

[0017] Determining the separated seismic P- and S-wave data of the survey area based on the three-component seismic data of the survey area and the P- and S-wave data separation model of the seismic area;

[0018] The seismic P-wave and S-wave data separation model is determined according to the seismic P-wave and S-wave data separation model construction method described in the first aspect.

[0019] In a fourth aspect, the present invention provides a seismic P-wave and S-wave data separation system, comprising:

[0020] A three-component seismic data acquisition module, used to acquire three-component seismic data of the survey area;

[0021] A seismic P- and S-wave data separation module, configured to determine the separated seismic P- and S-wave data of the survey area based on the three-component seismic data of the survey area and a seismic P- and S-wave data separation model;

[0022] The seismic P-wave and S-wave data separation model is determined according to the seismic P-wave and S-wave data separation model construction method described in the first aspect.

[0023] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0024] 1) Compared with conventional seismic data P-wave and S-wave separation methods, the present invention can obtain high-precision and high-resolution P-wave and S-wave data.

[0025] 2) The present invention adopts a data-driven workflow and intelligently separates longitudinal and shear wave data through artificial intelligence technology. It has high computational efficiency and does not rely on accurate medium parameters, thereby reducing work costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 It is a flow chart of the method for constructing a seismic P-wave and S-wave data separation model provided by the present invention;

[0028] Figure 2 Schematic diagram of the isotropic Marmousi-2 model provided by the present invention; Figure 2 (a) is a schematic diagram of the longitudinal wave velocity model. Figure 2 (b) is a schematic diagram of the shear wave velocity model;

[0029] Figure 3 The present invention uses the finite difference method to Figure 2 Schematic diagram of earthquake records obtained from the isotropic Marmousi-2 model: Figure 3(a) is a schematic diagram of the X component of the earthquake record obtained using the finite difference method. Figure 3 (b) is a schematic diagram of the Z component of the earthquake record obtained using the finite difference method. Figure 3 (c) is to use the present invention from Figure 3 Schematic diagram of the separated P-wave seismic records (a) and (b). Figure 3 (d) is to use the present invention from Figure 3 Schematic diagram of separated shear wave seismic records in (a) and (b);

[0030] Figure 4 The finite difference method is used to Figure 2 Snapshot diagram of the seismic wave field obtained from the isotropic Marmousi-2 model shown: Figure 4 (a) is a schematic diagram of the X component of the wave field snapshot obtained using the finite difference method. Figure 4 (b) is a schematic diagram of the Z component of the wave field snapshot obtained using the finite difference method;

[0031] Figure 5 The present invention is to use Figure 4 The separated longitudinal and transverse wave fields of (a) and (b) are: Figure 5 (a) is to use the present invention from Figure 4 Schematic diagram of the separated longitudinal wave fields (a) and (b). Figure 5 (b) is to use the present invention from Figure 4 Schematic diagram of the separated shear wave fields in (a) and (b);

[0032] Figure 6 is a schematic diagram of the anisotropic Marmousi-2 model provided by the present invention, Figure 6 (a) is the longitudinal wave velocity model v p Schematic diagram, Figure 6 (b) is the shear wave velocity model v S Schematic diagram, Figure 6 (c) is a schematic diagram of the first anisotropic parameter model ε, Figure 6 (d) is a schematic diagram of the second anisotropic parameter model δ, Figure 6 (e) is a schematic diagram of the structural dip model θ;

[0033] Figure 7 The finite difference method is used to Figure 6 Schematic diagram of earthquake records obtained from the anisotropic Marmousi-2 model: Figure 7 (a) is a schematic diagram of the X component of the earthquake record obtained using the finite difference method. Figure 7 (b) is a schematic diagram of the Z component of the earthquake record obtained using the finite difference method. Figure 7 (c) is to use the present invention from Figure 7 Schematic diagram of the separated P-wave seismic records (a) and (b). Figure 7 (d) is to use the present invention from Figure 7 Schematic diagram of separated P-wave seismic records in (a) and (b);

[0034] Figure 8 The finite difference method is used to Figure 6 A snapshot of the seismic wavefield obtained from the anisotropic Marmousi-2 model is shown: Figure 8 (a) is a schematic diagram of the X component of the wave field snapshot obtained using the finite difference method. Figure 8 (b) is a schematic diagram of the Z component of the wave field snapshot obtained using the finite difference method. Figure 8 (c) is to use the present invention from Figure 8 Schematic diagram of the separated longitudinal wave fields (a) and (b). Figure 8 (d) is to use the present invention from Figure 8 Schematic diagram of the separated shear wave fields in (a) and (b);

[0035] Figure 9 It is a flow chart of the method for separating seismic longitudinal and shear wave data provided by the present invention. DETAILED DESCRIPTION

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] Example 1

[0039] like Figure 1 As shown, this embodiment provides a method for constructing a model of seismic P- and S-wave data, comprising:

[0040] Step 100: Collect seismic data and construct several elastic models based on the seismic data. The elastic models are determined by a set of medium density, P-wave velocity, S-wave velocity, and anisotropy parameters. The seismic data includes at least geological profiles, gravity anomaly profiles, acoustic well logs, seismic imaging profiles, and seismic elastic interpretation profiles.

[0041] Step 100 specifically includes: first, establishing a large-scale geological and geophysical database based on the aforementioned seismic data; then, statistically analyzing the corresponding geophysical elastic characteristics, such as medium density, P-wave velocity, S-wave velocity, and anisotropy parameters, based on the large-scale geological and geophysical database; and then, constructing a large number of universal elastic models based on the geophysical elastic characteristics. For the purposes of this description, each elastic model is uniquely defined by a set of medium density, P-wave velocity, S-wave velocity, and anisotropy parameters.

[0042] Step 200: Process the elastic model to determine a sample data set; the sample data set includes multiple sample pairs; the sample pairs include sample input data and corresponding label data; the sample input data is three-component seismic data; the label data is separated P- and S-wave data.

[0043] As a preferred implementation of step 200, the step specifically includes:

[0044] S201: performing a transformation operation on the model parameters of the elastic model using a three-dimensional affine transformation algorithm to obtain a transformed elastic model.

[0045] For each of the elastic models, a three-dimensional affine transformation algorithm is used to perform transformation operations on the model parameters of the elastic model, thereby expanding more elastic models with rich and diverse characteristics.

[0046] The three-dimensional affine transformation algorithm is specifically a linear transformation algorithm from (x, y, z) coordinates to (u, v, w) coordinates, and its mathematical expression is expressed as:

[0047]

[0048] Among them, a i ,b i ,c i ,d i (i=1,2,3) is the transformation parameter in its corresponding transformation parameter matrix. The homogeneous coordinate matrix corresponding to formula (1) is:

[0049] U=AX (2);

[0050] In formula (2), U, X and A are specifically:

[0051] U=[uvw 1] T (3);

[0052] X=[xyz 1] T (4);

[0053]

[0054] In the above formula, the subscript "T" represents the matrix transpose operation; A represents the transformation parameter matrix corresponding to the affine transformation. Different matrices correspond to different affine transformations; affine transformation is implemented by one atomic transformation or a combination of several atomic transformations; atomic transformations specifically include translation, scaling, rotation, flipping, and shearing transformations.

[0055] The transformation matrix A corresponding to the translation transformation Trans Specifically:

[0056]

[0057] In formula (6), t x , t y and t z Represents the distance of translation along the coordinate axis x, y and z respectively.

[0058] The transformation matrix A corresponding to the scaling transformation Scale Specifically:

[0059]

[0060] In formula (7), s x , s y and s z Represents the scaling factors in the x, y, and z directions, respectively.

[0061] The transformation matrix A corresponding to the rotation transformation Rota Specifically:

[0062]

[0063] In formula (8), θ represents the rotation angle of the coordinate system, specifically the angle between the u direction of the transformed coordinate system and the x direction of the original coordinate system.

[0064] The transformation matrix A corresponding to the flip transformation Flip Specifically:

[0065]

[0066] In formula (9), κ x , κ y and κ z Represents the flip coefficients in the x, y and z directions of the coordinate axes respectively; when flipping along the x axis, κ x =-1,κ y =1,κ z =1; when flipping along the y-axis, κ x =1,κ y =-1,κ z =1; when flipping along the z axis, κx =1,κ y =1,κ z =-1.

[0067] The transformation matrix A corresponding to the shear transformation Shear Specifically:

[0068]

[0069] Formula (10) is the shearing along the x-direction, where α and β are the shearing angles between the v-direction and the w-direction of the transformed coordinate system and the y-direction and the z-direction of the original coordinate system, respectively.

[0070]

[0071] Formula (11) is the shear along the y direction, where φ and are the shear angles between the u and w directions of the transformed coordinate system and the x and z directions of the original coordinate system, respectively.

[0072]

[0073] Formula (12) is the shearing along the z direction, where ω and η are the shearing angles between the u and v directions of the transformed coordinate system and the x and y directions of the original coordinate system, respectively.

[0074] The three-dimensional affine transformation algorithm is used to perform transformation operations on the model parameters of the elastic model, thereby expanding more elastic models with rich and diverse features, wherein the transformation operations include three categories: the first type of transformation algorithm is an algorithm that applies a single atomic transformation in the three-dimensional affine transformation algorithm; the second type of transformation algorithm is an algorithm that applies a combination of multiple atomic transformations in the three-dimensional affine transformation algorithm; the third type of transformation algorithm is an algorithm that uses different atomic transformation parameters to perform transformations based on the second type of transformation algorithm, that is, applying the atomic transformation combination in the second type of transformation operation. In this example, the model parameters of the elastic model are transformed using the first type of transformation algorithm, the second type of transformation algorithm, and the third type of transformation algorithm in sequence, and finally a variety of elastic models with rich and diverse features are obtained.

[0075] S202: numerically solving the elastic wave equation of the anisotropic medium in the transformed elastic model using a finite difference method to obtain three-component seismic data.

[0076] Each elastic model obtained in step 201 is expanded, and based on the elastic wave equation of anisotropic medium, the finite difference method is used to numerically solve the elastic wave equation of anisotropic medium to obtain multiple groups of three-component seismic wave fields V(x)=(V x ,V y ,V z )T And the three-component seismic record d(x r ,t)=(d x ,d y ,d z ) T ; where x is the spatial position vector, x r is the spatial position vector of the detection point, the subscript "T" represents the matrix transpose operation, t represents time, V x , V y and V z They are the components of the three-component seismic wave field V(x) in the x, y and z directions, d x , d y and d z They are the three-component seismic records d(x r ,t) components in the x, y and z directions.

[0077] The elastic wave equation based on anisotropic media is specifically:

[0078]

[0079] In formula (13), C is the medium elastic stiffness matrix, L is the partial derivative operator matrix, ρ(x) is the medium density, V(x) is the elastic medium particle vibration velocity vector, that is, the three-component seismic wave field, σ(x) = (σ 11 ,σ 22 ,σ 33 ,σ 23 ,σ 31 ,σ 12 ) T is the stress tensor; the medium elastic stiffness matrix C is specifically:

[0080]

[0081] In formula (14), the relationship between the matrix elements and the elastic parameters is as follows:

[0082]

[0083] In formula (15), V P0 and V S0 are the vertical propagation velocities of longitudinal and shear waves, ε, δ, and γ are anisotropic parameters, and some stiffness matrix elements satisfy the following relationship: 22 =c 11 , c 55 =c 44 , c 12 =c 21 =c 11 -2c 66 , c 32 =c23 =c 13 =c 31 .

[0084] The partial derivative operator matrix L is specifically:

[0085]

[0086] The anisotropic medium elastic wave equation is numerically solved by the finite difference method based on the anisotropic medium elastic wave equation to obtain multiple groups of three-component seismic wave fields V(x) and three-component seismic records d(x) at different times. r ,t), specifically: using the staggered grid finite difference algorithm to solve the anisotropic medium elastic wave equation in the transformed elastic model, obtaining the medium particle vibration velocity vector field V(x), and saving the medium particle vibration velocity vector field V(x) at a given moment to determine it as the three-component seismic wave field at that moment until the maximum moment; at each moment of solving the three-component seismic wave field, extracting the corresponding vibration velocity vector field at the preset detector position, and determining this vibration velocity vector field as the seismic record at that position at that moment until the maximum moment, forming the three-component seismic record d(x r ,t).

[0087] Step 203: Solve the three-component seismic data based on a model-driven P-wave and S-wave separation algorithm to obtain separated P-wave and S-wave data.

[0088] Based on each set of three-component seismic wave fields V(x) obtained in step 202, the three-component seismic wave fields are solved using the model-driven seismic wave field longitudinal and transverse wave separation algorithm to obtain the seismic longitudinal and transverse wave fields V corresponding to the three-component seismic wave fields. m (x)=(V qP ,V qSV ,V qSH ) T Based on the obtained three-component seismic record d(x r ,t), solve the longitudinal and transverse wave seismic records using the model-driven wave field extension algorithm and the longitudinal and transverse wave separation algorithm of the seismic records, and obtain the longitudinal and transverse wave fields d corresponding to the three-component seismic records m (x r ,t)=(d qP ,d qSV ,d qSH ) T ; Among them, V qP , V qSV and V qSH The separated longitudinal and transverse wave fields V m (x) qP, qSV and qSH wave field components, d qP , d qSV and d qSHrepresents the separated P-wave and S-wave seismic records d m (x r The qP, qSV and qSH wave components of the seismic record are shown in Figure 1. The superscript "m" indicates the wave mode, specifically including qP, qSV and qSH. The seismic P- and S-wave fields are used to represent the separated P- and S-wave data.

[0089] The method is based on each set of three-component seismic wave fields V(x) and uses the model-driven seismic wave field longitudinal and transverse wave separation algorithm to solve the longitudinal and transverse wave fields V m (x), specifically, the longitudinal and transverse wave separation equations of the seismic wave field are used to realize the longitudinal and transverse wave separation, and the equation is:

[0090]

[0091] In formula (17), F[·] represents Fourier transform, F -1 [·] represents the inverse Fourier transform, a qP is the qP wave polarization vector, a qSV is the polarization vector of qSV wave, a qSH is the polarization vector of qSH wave, and “i” is the imaginary unit.

[0092] The three-component seismic record d(x r ,t), using the model-driven wave field extension algorithm and seismic record P-wave and S-wave separation algorithm to solve the P-wave and S-wave seismic record d m (x r ,t), the specific process is:

[0093] (a) A near-surface velocity model and anisotropy parameter model are established. Then, using the multi-component seismic records as boundary conditions, reverse-time extension of the elastic wavefield is performed, extending the seismic records from the geophone acquisition plane to a reference depth underground. The equation used for elastic wavefield extension is:

[0094]

[0095] In formula (18), c ijkl is the elastic stiffness matrix of the medium, where the subscripts i, j, k, and l range from x, y, and z, respectively representing the directions of the three coordinate axes x, y, and z in the Cartesian coordinate system;

[0096] (b) Using the elastic wave field extended from step (a) above, perform P-wave and S-wave separation of the wave field snapshot based on formula (17), and extract the P-wave and S-wave seismic records on the reference surface. in, and Represents the P-wave and S-wave seismic records on the reference surface The qP, qSV and qSH components of the earthquake record, the subscript "Ref" indicates the reference surface.

[0097] (c) Seismic records of longitudinal and transverse waves on the reference surface Each component is a boundary value condition, and the scalar wave field is forward extended based on the pure wave equation. At each time step of the scalar wave field forward extension, the separated P-wave and S-wave seismic records d are extracted at the acquisition surface where the original geophone is located. m (x r ,t); wherein the pure wave equations include two, namely the pure qP wave equation and the pure qS (including qSV and qSH) wave equation.

[0098] The pure qP wave equation is specifically:

[0099]

[0100] The pure qS wave equation is specifically:

[0101]

[0102] In the above formula, represents the horizontal Laplace operation, v p0 and v s0 is the velocity of qP and qS waves along the symmetry axis; is the horizontal velocity of the qP wave; the operator R m The specific expression is as follows:

[0103]

[0104] In formula (21),

[0105] Step 300: Using the sample data set to train the artificial intelligence network, a seismic P-wave and S-wave data separation model is obtained; the seismic P-wave and S-wave data separation model is used to separate the seismic P-wave and S-wave data in the survey area.

[0106] The three-component seismic wave field V(x) obtained in step 200 is used to construct the sample input data required for the seismic wave field snapshot intelligent separation network training, and the longitudinal and transverse wave fields V m (x) is used as the real longitudinal and transverse wave field data required for network training, namely the label data; the obtained three-component seismic record d(x r ,t) Construct the sample input data required for the training of the seismic record intelligent separation network, and use the obtained longitudinal and transverse wave seismic records d m (x r ,t) is the real P- and S-wave seismic recording data required for network training, namely the label data.

[0107] The WcycleGAN artificial intelligence network consists of two groups of generators G MN , generator G NM and two sets of discriminators D M , Discriminator D N ; The generator G MN and generator G NM Composed of TransUNet network, the discriminator D M and the discriminator D N It consists of multiple convolutional blocks.

[0108] The WcycleGAN artificial intelligence network is specifically an artificial intelligence network formed by introducing the Wasserstein distance W and the gradient penalty mechanism GP on the basis of the CycleGAN artificial intelligence network.

[0109] The Wasserstein distance W is specifically:

[0110]

[0111] In formula (22), Π(·,·) represents the set of joint distributions, P r and P g Denotes Π(P r ,P g ) in each distribution, ξ represents the marginal distribution of Π(P r ,P g ), e is the real sample, o is the generated sample, E (e,o)～ξ Represents the expected value of the sample in the joint distribution ξ, and inf represents the lower bound of the function.

[0112] The gradient penalty mechanism specifically adds a gradient penalty term to the loss function in the artificial intelligence network; the gradient penalty term is specifically the Lipschitz condition, which uses this condition to constrain the artificial intelligence network so that the gradient of the loss function is less than or equal to 1 when the gradient of the loss function is greater than 1.

[0113] The WcycleGAN artificial intelligence network is trained using the sample data set, and the G in WcycleGAN is continuously optimized through a large number of rounds of training. MN , G NM 、D M and D N The parameters of the training data are dynamically adjusted using an adaptive learning rate to form a seismic P-wave and S-wave data separation model.

[0114] The training process of the WcycleGAN artificial intelligence network is as follows: during the training process, the WcycleGAN network is divided into a Data Cycle part and a Model Cycle part; in the Data Cycle part, the three-component seismic wave field V(x) (or d(x)) in the sample data set is trained. r ,t)) is fed into the generator G as the input of the training network MN In, through G MN Output predicted P-wave and S-wave data (or ), the predicted longitudinal and transverse wave data V m_pre (x)(or d m_pre (x r ,t)) is sent to the discriminator D N In the discriminator D N The Wasserstein distance W and the gradient penalty mechanism GP are used to calculate the adversarial loss of the Data Cycle part. Then the predicted longitudinal and transverse wave data V m_pre (x)(or d m_pre (x r ,t)) is fed into the generator G NM Reconstructing three-component seismic wave fields (or ), and calculate the cycle consistency loss of the Data Cycle part at the same time Based on the optimization inversion algorithm tends to be minimum; the superscript "m_pre" represents the predicted P-wave and S-wave data, the superscript "rc" represents the reconstructed three-component seismic data, and the superscript "d" represents the DataCycle part of the network training.

[0115] The adversarial loss in the Data Cycle part Specifically:

[0116]

[0117] Wherein, the hyperparameter λ1 represents the regularization factor, is the loss function of the Wasserstein distance in the Data Cycle part, It is the loss function of GP in Data Cycle.

[0118] The cycle consistency loss of the Data Cycle part Specifically:

[0119]

[0120] Where M represents the labeled three-component seismic data, M * represents the unlabeled three-component seismic data, W * represents the trainable parameters in the neural network, Represents a neural network operator that generates predictions based on input data.

[0121] In the Model Cycle section, the longitudinal and transverse wave fields V in the sample data set are m (x)(or d m (x r ,t)) is fed into the generator G as the input of the training network NM In, through G NM Output predicted three-component seismic wave field (or ), the predicted three-component earthquake data V pre (x)(or d pre (x r ,t)) is sent to the discriminator D M In the discriminator D M The Wasserstein distance W and the gradient penalty mechanism GP are used to calculate the adversarial loss of the Model Cycle part. Then the predicted three-component seismic data V pre (x)(or d pre (x r ,t)) is fed into the generator G MN Reconstructing P- and S-wave data (or ), calculate the cycle consistency loss of the ModelCycle part Based on the optimization inversion algorithm tends to be minimum; the superscript "pre" represents the predicted three-component seismic data, the superscript "m_rc" represents the reconstructed P- and S-wave data, and the superscript Indicates the Model Cycle part.

[0122] The adversarial loss of the Model Cycle part Specifically:

[0123]

[0124] Where, the hyperparameter λ2 represents the regularization factor, is the loss function of the Wasserstein distance of the Model Cycle part, It is the loss function of GP in Model Cycle;

[0125] The cycle consistency loss of the Model Cycle part Specifically:

[0126]

[0127] Where N represents the longitudinal and transverse wave data, W * represents the trainable parameters in the neural network, Represents a neural network operator that generates predictions based on input data.

[0128] The adversarial loss L of the WcycleGAN network adv It is the sum of the Data Cycle part and the Model Cycle part, specifically:

[0129]

[0130] The cycle consistency loss L of the WcycleGAN network cyc is the sum of the Data Cycle part and the Model Cycle part, specifically:

[0131]

[0132] The loss function L of the entire WcycleGAN is expressed as:

[0133] L=L adv +λL cyc (29).

[0134] Where, the hyperparameter λ represents the regularization factor;

[0135] The adaptive learning rate algorithm is specifically as follows: in the first 100 rounds, the learning rate lr is fixed at 0.0002, and in the next 100 rounds, the learning rate change function lr is set to i =0.95*lr i-1 .

[0136] This embodiment relates to a method for constructing a seismic P-wave and S-wave data separation model. By analyzing the geophysical elastic characteristics in geological profiles, gravity anomaly profiles, acoustic logging curves, seismic imaging profiles, and seismic elastic interpretation profiles, a large number of universal elastic models are constructed, and the diversity and number of elastic models are increased through affine transformation. Based on the elastic wave equation of anisotropic media, the finite difference method is used to numerically solve the elastic wave equation to obtain three-component seismic data. The three-component seismic data is solved based on a model-driven P-wave and S-wave separation method to obtain separated P-wave and S-wave data. The three-component seismic data and P-wave and S-wave data are used as data sets and trained using the WCycleGAN artificial intelligence network to obtain an intelligent P-wave and S-wave separation method for P-wave and S-wave separation. The present invention introduces artificial intelligence technology into the process of seismic P-wave and S-wave separation. By establishing a mapping relationship between the three-component seismic data and P-wave and S-wave data, P-wave and S-wave are extracted from the three-component seismic data to obtain high-precision P-wave and S-wave data. The P-wave and S-wave data obtained by this method meet the accuracy requirements, have high computational efficiency, do not rely on accurate medium parameters, and can be applied to various complex geological conditions, laying a solid data foundation for actual seismic data inversion imaging.

[0137] To further illustrate the feasibility and effectiveness of the present invention, two examples are given below:

[0138] Example 1:

[0139] like Figure 2 As shown, the isotropic Marmousi-2 model contains 801×561 grids, the spatial interval in each direction is 10m, the earthquake source is located at (4000m, 100m), the time sampling interval is 1ms, an explosion source is used to excite seismic waves, and the wavelet used is a Ricker wavelet with a main frequency of 15Hz. Figure 3 The finite difference method is used to Figure 2 The seismic records obtained from the isotropic Marmousi-2 model are shown. Figure 3 As can be seen from (c) and (d) in the figure, the separation effect of the longitudinal and transverse wave seismic records obtained by the method of the present invention is good, which proves the feasibility and effectiveness of the method of the present invention. Figure 4 The finite difference method is used to Figure 2 A snapshot of the seismic wavefield obtained from the isotropic Marmousi-2 model is shown. Figure 5 The present invention is to use Figure 4 The longitudinal and transverse wave fields separated by (a) and (b). Figure 4 and Figure 5 It can be seen that the separated wave field obtained by the method of the present invention accurately maintains the amplitude and phase characteristics of the wave field, and the longitudinal and transverse waves are separated cleanly, which proves the effectiveness of the method of the present invention.

[0140] Example 2:

[0141] Figure 6 It is the anisotropic Marmousi-2 model provided by the present invention. The model contains 801×541 grids, the spatial interval in each direction is 10m, a vertical concentrated force source is used to excite seismic waves, the wavelet used is a Ricker wavelet with a main frequency of 20Hz, the source is located at (4000m, 100m), and the time sampling interval is 1ms. Figure 7 The finite difference method is used to Figure 6 The seismic records obtained from the anisotropic Marmousi-2 model are shown in Figure 1. Figure 7 It can be seen from the seismic records shown that for complex models, the longitudinal and transverse waves are well separated by the longitudinal and transverse wave seismic records obtained by the present invention, which shows the effectiveness of the method of the present invention. Figure 8 The finite difference method is used to Figure 6 Snapshot of the seismic wavefield obtained from the anisotropic Marmousi-2 model shown. Figure 8 As can be seen, the P-wave and S-wave separation method proposed in this invention can still accurately separate the P-wave and S-wave fields in complex models. These results demonstrate the effectiveness of the method in complex models. In summary, the method proposed in this invention has excellent feasibility and practicality in complex geological and geophysical models.

[0142] Example 2

[0143] In order to execute the corresponding method of the above-mentioned embodiment 1 and realize the corresponding functions and technical effects, a system for constructing a seismic P-wave and S-wave data separation model is provided below.

[0144] This embodiment provides a seismic P-wave and S-wave data separation model construction system comprising:

[0145] The elastic model building module is used to collect seismic data and build several elastic models based on the seismic data; the elastic model is determined by a set of medium density, longitudinal wave velocity, shear wave velocity and anisotropy parameters.

[0146] A sample data set determination module is used to process the elastic model and determine a sample data set; the sample data set includes multiple sample pairs; the sample pairs include sample input data and corresponding label data; the sample input data is three-component seismic data; the label data is separated longitudinal and transverse wave data.

[0147] The earthquake P-wave and S-wave data separation model construction module is used to train the artificial intelligence network using a sample data set to obtain an earthquake P-wave and S-wave data separation model; the earthquake P-wave and S-wave data separation model is used to separate the earthquake P-wave and S-wave data in the survey area.

[0148] Example 3

[0149] like Figure 9 As shown, this embodiment provides a method for separating seismic P- and S-wave data, comprising:

[0150] Step 901: Acquire three-component seismic data of the survey area.

[0151] Step 902: Determine the P- and S-wave data separated from the survey area based on the three-component seismic data and the P- and S-wave data separation model of the survey area. The P- and S-wave data separation model is determined by the P- and S-wave data separation model construction method described in Example 1.

[0152] Accordingly, this embodiment provides a seismic P-wave and S-wave data separation system, comprising:

[0153] The three-component seismic data acquisition module is used to acquire three-component seismic data of the survey area.

[0154] The seismic P- and S-wave data separation module is used to determine the separated seismic P- and S-wave data in the survey area based on the three-component seismic data in the survey area and the seismic P- and S-wave data separation model.

[0155] The seismic P-wave and S-wave data separation model is determined according to the seismic P-wave and S-wave data separation model construction method described in Example 1.

[0156] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0157] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for constructing a seismic P-wave and S-wave data separation model, characterized in that: include: Collecting seismic data and constructing a plurality of elastic models based on the seismic data; the elastic models are determined by a set of medium density, longitudinal wave velocity, shear wave velocity and anisotropy parameters; The elastic model is processed to determine a sample data set; the sample data set includes a plurality of sample pairs; the sample pairs include sample input data and corresponding label data; the sample input data is three-component seismic data; the label data is separated longitudinal and transverse wave data; The artificial intelligence network is trained using the sample data set to obtain a seismic P-wave and S-wave data separation model; The seismic P-wave and S-wave data separation model is used to separate the seismic P-wave and S-wave data in the survey area; The processing of the elasticity model to determine the sample data set specifically includes: Performing a transformation operation on the model parameters of the elastic model using a three-dimensional affine transformation algorithm to obtain a transformed elastic model; Numerically solving the elastic wave equation of the anisotropic medium in the transformed elastic model using a finite difference method to obtain three-component seismic data; the three-component seismic data includes multiple groups of three-component seismic wave fields at different times and corresponding three-component seismic records; Solving the three-component seismic data based on a model-driven P-wave and S-wave separation algorithm to obtain separated P-wave and S-wave data, wherein the separated P-wave and S-wave data include P-wave and S-wave fields and P-wave and S-wave seismic records; The three-dimensional affine transformation algorithm is used to transform the model parameters of the elastic model, specifically including: sequentially using a first type of transformation algorithm, a second type of transformation algorithm, and a third type of transformation algorithm to perform transformation operations on the model parameters of the elastic model to obtain a transformed elastic model; The first type of transformation algorithm is an algorithm for applying a single atomic transformation in a three-dimensional affine transformation algorithm; The second type of transformation algorithm is an algorithm that applies a combination of multiple atomic transformations in a three-dimensional affine transformation algorithm; The third type of transformation algorithm is an algorithm that uses different atomic transformation parameters to perform transformation based on the second type of transformation algorithm; Using three-component seismic wave fields as input data, using longitudinal and transverse wave fields as real longitudinal and transverse wave seismic wave field data, i.e., label data; using three-component seismic records as input data, using longitudinal and transverse wave seismic records as real longitudinal and transverse wave seismic record data, i.e., label data; The artificial intelligence network is trained using the sample data set to obtain a seismic P-wave and S-wave data separation model, specifically including: Constructing a WcycleGAN artificial intelligence network; the WcycleGAN artificial intelligence network is an artificial intelligence network formed by introducing the Wasserstein distance W and the gradient penalty mechanism GP on the basis of the CycleGAN artificial intelligence network; The WcycleGAN artificial intelligence network is trained using a sample data set to obtain a seismic P-wave and S-wave data separation model.

2. A method for constructing a seismic P-wave and S-wave data separation model according to claim 1, characterized in that: The seismic data at least includes geological sections, gravity anomaly sections, acoustic logging curves, seismic imaging sections and seismic elastic interpretation sections; the construction of several elastic models based on the seismic data specifically includes: Building a geological and geophysical database based on the seismic data; determining geophysical elastic characteristics based on the geological and geophysical database; Based on the geophysical elastic characteristics, several elastic models are constructed.

3. A method for constructing a seismic P-wave and S-wave data separation model according to claim 1, characterized in that: The method of numerically solving the elastic wave equation of the anisotropic medium in the transformed elastic model using the finite difference method to obtain three-component seismic data specifically includes: Using a staggered grid finite difference algorithm to solve the anisotropic medium elastic wave equation in the transformed elastic model, obtaining a medium particle vibration velocity vector field, and determining the medium particle vibration velocity vector field saved at a given time as a three-component seismic wave field at the given time until a maximum time; the given time is any time when solving the three-component seismic wave field; Then, the three-component seismic wave field is solved for each moment, and the corresponding vibration velocity vector field extracted at the geophone position is determined as the three-component seismic record at the given moment, until the maximum moment.

4. A method for constructing a seismic P-wave and S-wave data separation model according to claim 1, characterized in that: The model-driven P-wave and S-wave separation algorithm solves the three-component seismic data to obtain separated P-wave and S-wave data, specifically including: Solving the three-component seismic wave field using a model-driven longitudinal and transverse wave separation algorithm to obtain longitudinal and transverse wave fields corresponding to the three-component seismic wave field; Solving the three-component seismic record using a model-driven wavefield continuation algorithm and a seismic record longitudinal and transverse wave separation algorithm to obtain seismic longitudinal and transverse wavefields corresponding to the three-component seismic record; The seismic P- and S-wave fields are used to represent the separated P- and S-wave data.

5. A seismic P-wave and S-wave data separation model construction system, characterized in that: include: An elastic model building module, for collecting seismic data and building a plurality of elastic models based on the seismic data; the elastic models are determined by a set of medium density, longitudinal wave velocity, shear wave velocity and anisotropy parameters; A sample data set determination module is used to process the elastic model to determine a sample data set; the sample data set includes a plurality of sample pairs; the sample pairs include sample input data and corresponding label data; the sample input data is three-component seismic data; the label data is separated P-wave and S-wave data; Seismic P-wave and S-wave data separation model construction module, used to train the artificial intelligence network using sample data sets to obtain a seismic P-wave and S-wave data separation model; The seismic P-wave and S-wave data separation model is used to separate the seismic P-wave and S-wave data in the survey area; The processing of the elasticity model to determine the sample data set specifically includes: Performing a transformation operation on the model parameters of the elastic model using a three-dimensional affine transformation algorithm to obtain a transformed elastic model; Numerically solving the elastic wave equation of the anisotropic medium in the transformed elastic model using a finite difference method to obtain three-component seismic data; the three-component seismic data includes multiple groups of three-component seismic wave fields at different times and corresponding three-component seismic records; Solving the three-component seismic data based on a model-driven P-wave and S-wave separation algorithm to obtain separated P-wave and S-wave data, wherein the separated P-wave and S-wave data include P-wave and S-wave fields and P-wave and S-wave seismic records; The three-dimensional affine transformation algorithm is used to transform the model parameters of the elastic model, specifically including: sequentially using a first type of transformation algorithm, a second type of transformation algorithm, and a third type of transformation algorithm to perform transformation operations on the model parameters of the elastic model to obtain a transformed elastic model; The first type of transformation algorithm is an algorithm for applying a single atomic transformation in a three-dimensional affine transformation algorithm; The second type of transformation algorithm is an algorithm that applies a combination of multiple atomic transformations in a three-dimensional affine transformation algorithm; The third type of transformation algorithm is an algorithm that uses different atomic transformation parameters to perform transformation based on the second type of transformation algorithm; Using three-component seismic wave fields as input data, using longitudinal and transverse wave fields as real longitudinal and transverse wave seismic wave field data, i.e., label data; using three-component seismic records as input data, using longitudinal and transverse wave seismic records as real longitudinal and transverse wave seismic record data, i.e., label data; The artificial intelligence network is trained using the sample data set to obtain a seismic P-wave and S-wave data separation model, specifically including: Constructing a WcycleGAN artificial intelligence network; the WcycleGAN artificial intelligence network is an artificial intelligence network formed by introducing the Wasserstein distance W and the gradient penalty mechanism GP on the basis of the CycleGAN artificial intelligence network; The WcycleGAN artificial intelligence network is trained using a sample data set to obtain a seismic P-wave and S-wave data separation model.

6. A method for separating seismic P-wave and S-wave data, characterized in that: include: Acquire three-component seismic data of the survey area; Determining the separated seismic P- and S-wave data of the survey area based on the three-component seismic data of the survey area and the P- and S-wave data separation model of the seismic area; The seismic P-wave and S-wave data separation model is determined according to the seismic P-wave and S-wave data separation model construction method according to any one of claims 1 to 4.

7. A seismic P-wave and S-wave data separation system, characterized in that: include: A three-component seismic data acquisition module, used to acquire three-component seismic data of the survey area; A seismic P- and S-wave data separation module, configured to determine the separated seismic P- and S-wave data of the survey area based on the three-component seismic data of the survey area and a seismic P- and S-wave data separation model; The seismic P-wave and S-wave data separation model is determined according to the seismic P-wave and S-wave data separation model construction method according to any one of claims 1 to 4.

Images