Elastic wave full waveform inversion method, system, device and medium

Abstract

The application relates to the technical field of elastic waves, and discloses an elastic wave full waveform inversion method, which comprises the following steps: inputting observed seismic data into a preset neural network encoder to extract high-dimensional feature representation; inputting the high-dimensional feature representation into a preset neural network decoder to obtain a gradient update value; obtaining an elastic parameter model based on the gradient update value and an initial elastic parameter model; obtaining Lame parameters based on P-wave velocity, S-wave velocity and medium density; obtaining synthetic seismic data based on the Lame parameters and an elastic wave equation; obtaining a target function based on the observed seismic data and the synthetic seismic data; iteratively updating preset neural network weights by using the target function until a loss function reaches a first preset threshold or the number of iterations reaches a second preset threshold, so that a target gradient update value is obtained; and obtaining a final target elastic parameter model based on the target gradient update value and a target elastic parameter model, so as to relieve the ill-posedness caused by simultaneous inversion of multiple elastic parameters.

Description

Technical Field

[0001] This application relates to the field of elastic wave technology, specifically to an elastic wave full waveform inversion method, system, device, and medium. Background Technology

[0002] With the continuous advancement of oil and gas exploration and development technologies, constructing high-precision subsurface parameter models can provide reliable elastic parameters for subsequent oil and gas interpretation work, which is of great significance for seismic imaging, geological interpretation, and oil and gas exploration and development. Full-waveform inversion methods, due to their ability to fully utilize the kinematic and dynamic information of seismic waves, have gradually become a hot research topic. Deep learning methods have been gradually introduced into the elastic wave full-waveform inversion process and have become one of the commonly used technical approaches.

[0003] However, traditional acoustic wave full waveform inversion is typically ill-posed and prone to cycle skipping. Furthermore, the joint inversion of multiple elastic parameters (P-wave velocity, S-wave velocity, and medium density) and multi-component data processing in elastic wave full waveform inversion further exacerbate the ill-posedness, making it difficult to obtain a satisfactory inversion parameter model. Especially when the initial elastic parameter model has biases, the inversion process is highly susceptible to cycle skipping, hindering high-precision modeling and often resulting in strong ill-posedness and severe parameter gradient coupling. Summary of the Invention

[0004] The purpose of this application is to provide a method, system, device and medium for elastic wave full waveform inversion, in order to solve the problems in the prior art where the inversion process is prone to cycle skipping, making it difficult to achieve high-precision modeling, and often facing strong ill-stability and severe parameter gradient coupling.

[0005] To achieve the above objectives, the first aspect of this application provides a method for full waveform inversion of elastic waves, the method comprising:

[0006] Acquire initial synthetic seismic data, observed seismic data, and a first initial elastic parameter model. The observed seismic data includes the horizontal component and the vertical component of the initial particle velocity vector.

[0007] The observed seismic data is input into the encoder module of a pre-defined neural network to extract high-dimensional feature representations of the observed seismic data. The pre-defined neural network is trained using the Unet model.

[0008] The high-dimensional feature representation is input into the decoder module of the pre-defined neural network to obtain the initial gradient update value of the elasticity parameter;

[0009] Based on the initial gradient update value and the first initial elastic parameter model, the first target elastic parameter model related to the longitudinal wave velocity, transverse wave velocity and medium density is obtained.

[0010] Based on the longitudinal wave velocity, transverse wave velocity, and medium density, the Lamé parameters of the target medium are obtained;

[0011] Based on the Lamé parameters of the target medium and the elastic wave equation, synthetic multi-component seismic data is obtained, which includes the horizontal component and the vertical component of the target particle velocity vector.

[0012] The objective function is derived based on observed seismic data and synthetic multi-component seismic data;

[0013] The weights of the preset neural network are iteratively updated using the objective function until the error obtained based on the observed seismic data, the synthesized multi-component seismic data and the loss function reaches the first preset threshold or the number of iterations reaches the second preset threshold, thereby obtaining the target gradient update value of the elastic parameter.

[0014] Based on the target gradient update value and the second target elastic parameter model, the final target elastic parameter model related to the target P-wave velocity, the target S-wave velocity and the target medium density is obtained. The second target elastic parameter model is determined based on the gradient update value and elastic parameter model obtained by the iterative update of the weights of the previous preset neural network.

[0015] In this embodiment of the application, the first target elasticity parameter model includes:

[0016]

[0017] In the formula, This represents the first objective elasticity parameter model. This indicates the decoder module. Indicates encoder module, Indicates network weights, This indicates observed earthquake data. Indicates custom weights. Indicates the update step size. This represents the first initial elasticity parameter model.

[0018] In this embodiment of the application, the Lamé parameters of the target medium are obtained based on the longitudinal wave velocity, the transverse wave velocity, and the medium density, including:

[0019] Based on the P-wave velocity, S-wave velocity, and medium density, as well as the Lamé parameter calculation formula, the Lamé parameters of the target medium are obtained. The Lamé parameter calculation formula includes:

[0020]

[0021]

[0022] In the formula, and All represent the Lamé parameters of the target medium. Indicates the density of the medium. Indicates the transverse wave velocity. This indicates the longitudinal wave velocity.

[0023] In this embodiment of the application, obtaining initial synthetic seismic data includes:

[0024] Obtain the initial medium density, initial longitudinal wave velocity, and initial transverse wave velocity;

[0025] Based on the initial medium density, initial longitudinal wave velocity, and initial transverse wave velocity, the initial Lamé parameters of the medium are obtained;

[0026] Initial synthetic seismic data are obtained based on the initial medium Lamé parameters and elastic wave equations.

[0027] In this embodiment of the application, the elastic wave equation includes:

[0028]

[0029]

[0030]

[0031]

[0032]

[0033] In the formula, Indicates the density of the medium. Indicates the horizontal plane. Indicates a vertical plane. The horizontal component of the particle velocity vector. Indicates the propagation time of seismic waves. express Stress response in the direction, express Stress response in the direction, The vertical plane component representing the particle velocity vector. express Stress response in the direction, Indicates the media Lame parameters, This indicates the media lamellar parameters.

[0034] In this embodiment, the encoder module includes an encoder, which is composed of multiple alternating 3*3 convolutional layers and 2*2 max pooling layers.

[0035] In this embodiment, the decoder module includes a first decoder, a second decoder, and a third decoder. The first decoder is used to generate longitudinal wave velocity, the second decoder is used to generate transverse wave velocity, and the third decoder is used to generate dielectric density. The first decoder, the second decoder, and the third decoder are all composed of multiple 2*2 deconvolution layers.

[0036] A second aspect of this application provides a full waveform inversion system for elastic waves, the system comprising:

[0037] The acquisition module is used to acquire initial synthetic seismic data, observed seismic data, and a first initial elastic parameter model. The observed seismic data includes the horizontal component and the vertical component of the initial particle velocity vector.

[0038] The extraction module is used to input the observed seismic data into the encoder module of the preset neural network to extract the high-dimensional feature representation of the observed seismic data. The preset neural network is trained through the Unet model.

[0039] The first module is used to input the high-dimensional feature representation into the decoder module of the preset neural network to obtain the initial gradient update value of the elasticity parameter.

[0040] The second module is used to obtain a first target elastic parameter model related to longitudinal wave velocity, transverse wave velocity and medium density based on the initial gradient update value and the first initial elastic parameter model.

[0041] The third module is used to obtain the Lamé parameters of the target medium based on the longitudinal wave velocity, transverse wave velocity, and medium density.

[0042] The fourth module is used to obtain synthetic multi-component seismic data based on the Lamé parameters of the target medium and the elastic wave equation. The synthetic multi-component seismic data includes the horizontal component and the vertical component of the target particle velocity vector.

[0043] The fifth module is used to obtain the objective function based on observed seismic data and synthetic multi-component seismic data;

[0044] The iterative update module is used to iteratively update the weights of the preset neural network using the objective function until the error obtained based on the observed seismic data, the synthesized multi-component seismic data and the loss function reaches the first preset threshold or the number of iterations reaches the second preset threshold, so as to obtain the target gradient update value of the elastic parameter.

[0045] The sixth module is used to obtain the final target elastic parameter model related to the target longitudinal wave velocity, the target transverse wave velocity, and the target medium density based on the target gradient update value and the second target elastic parameter model. The second target elastic parameter model is determined based on the gradient update value and elastic parameter model obtained by the iterative update of the weights of the previous preset neural network.

[0046] A third aspect of this application provides an elastic wave full waveform inversion device, comprising:

[0047] The memory is configured to store instructions;

[0048] The processor is configured to retrieve instructions from memory and, when executing instructions, to implement the elastic wave full waveform inversion method as described in the first aspect above.

[0049] A fourth aspect of this application provides a machine-readable storage medium storing instructions for causing a machine to perform an elastic wave full waveform inversion method as described in the first aspect above.

[0050] The above technical solutions effectively alleviate the problem of local artifacts in the results caused by the limited expressive power of neural networks and the ill-posedness problem caused by gradient crosstalk between different elastic parameters during the joint inversion of multiple elastic parameters.

[0051] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0052] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings:

[0053] Figure 1 The schematic diagram illustrates a flow chart of an elastic wave full waveform inversion method according to an embodiment of this application;

[0054] Figure 2 The illustration shows a schematic diagram of a real elastic parameter model based on a two-dimensional elastic Marmousi model simulation according to an embodiment of this application;

[0055] Figure 3 The illustration shows a schematic diagram of an initial elasticity parameter model based on an embodiment of this application;

[0056] Figure 4 The illustration shows a schematic diagram of an elastic parameter model obtained by the elastic wave full waveform inversion method according to an embodiment of the present application;

[0057] Figure 5 The illustration shows a schematic diagram of an elastic parameter model obtained by an elastic wave full waveform inversion method based on conventional physics-driven methods according to an embodiment of this application.

[0058] Figure 6 A schematic diagram illustrating the comparison of single-track curves of elastic parameters according to embodiments of this application is provided.

[0059] Figure 7 The illustration shows a comparative diagram of the changing trend of a normalized loss function according to an embodiment of this application;

[0060] Figure 8 The illustration shows a schematic diagram of a real elastic parameter model based on a two-dimensional elastic BP model according to an embodiment of this application;

[0061] Figure 9 A schematic diagram of another initial elasticity parameter model according to an embodiment of this application is shown;

[0062] Figure 10 The illustration shows a schematic diagram of an elastic parameter model obtained by another elastic wave full waveform inversion method based on the present application, according to an embodiment of the present application;

[0063] Figure 11 The illustration shows a schematic diagram of an elastic parameter model obtained by another elastic wave full waveform inversion method based on conventional physical drive according to an embodiment of this application;

[0064] Figure 12 The illustration shows a comparative diagram of the changing trend of another normalized loss function according to an embodiment of this application. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0066] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0067] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0068] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0069] Figure 1 The illustration schematically shows a flowchart of an elastic wave full waveform inversion method according to an embodiment of this application. Figure 1 As shown in the embodiments of this application, a method for full waveform inversion of elastic waves is provided, which may include the following steps:

[0070] Step S110: Acquire initial synthetic seismic data, observed seismic data, and a first initial elastic parameter model, wherein the observed seismic data includes the horizontal component and the vertical component of the initial particle velocity vector;

[0071] Step S120: Input the observed seismic data into the encoder module of the preset neural network to extract the high-dimensional feature representation of the observed seismic data. The preset neural network is trained using the Unet model.

[0072] Step S130: Input the high-dimensional feature representation into the decoder module of the preset neural network to obtain the initial gradient update value of the elasticity parameter;

[0073] Step S140: Based on the initial gradient update value and the first initial elastic parameter model, obtain the first target elastic parameter model related to the longitudinal wave velocity, transverse wave velocity and medium density;

[0074] Step S150: Based on the longitudinal wave velocity, transverse wave velocity, and medium density, obtain the Lamé parameters of the target medium;

[0075] Step S160: Based on the Lamé parameters of the target medium and the elastic wave equation, synthesized multi-component seismic data is obtained, wherein the synthesized multi-component seismic data includes the horizontal component and the vertical component of the target particle velocity vector.

[0076] Step S170: Based on observed seismic data and synthetic multi-component seismic data, obtain the objective function;

[0077] Step S180: Iteratively update the weights of the preset neural network using the objective function until the error obtained based on the observed seismic data, the synthesized multi-component seismic data and the loss function reaches the first preset threshold or the number of iterations reaches the second preset threshold, and obtain the target gradient update value of the elastic parameter.

[0078] Step S190: Based on the target gradient update value and the second target elastic parameter model, obtain the final target elastic parameter model related to the target longitudinal wave velocity, the target transverse wave velocity and the target medium density. The second target elastic parameter model is determined based on the gradient update value and elastic parameter model obtained by the iterative update of the weights of the previous preset neural network.

[0079] In step S110, the initial synthetic seismic data is seismic wave data artificially generated by theoretical models or numerical simulation methods, and the observed seismic data is seismic wave data acquired by actual seismographs. The first initial elastic parameter model is a mathematical framework used in seismology to describe the elastic properties of a medium. Its core is to quantify the medium's response to elastic waves through a set of physical parameters.

[0080] In step S120, the observed seismic data is input into the encoder module of a preset neural network trained by the Unet model to extract high-dimensional feature representations of the observed seismic data.

[0081] In step S130, the high-dimensional feature representation of the observed seismic data is input into the decoder module of the preset neural network trained by the Unet model to obtain the initial gradient update value of the elastic parameters.

[0082] In step S140, based on the initial gradient update values ​​of the elastic parameters obtained in step S130 and the first initial elastic parameter model obtained in step S110, the first target elastic parameter model related to the P-wave velocity, S-wave velocity, and medium density can be obtained. Here, the P-wave velocity, S-wave velocity, and medium density are all elastic parameters.

[0083] In step S150, the target medium Lamé parameters are obtained by converting the longitudinal wave velocity, transverse wave velocity, and medium density.

[0084] In step S160, the Lamé parameters of the target medium are substituted into the elastic wave equation for calculation to obtain the synthetic multi-component seismic data.

[0085] In step S170, in the conventional elastic wave full-waveform inversion of existing technologies, the objective function typically uses the L2 norm to measure the residual between the observed seismic data and the forward modeling data. This embodiment of the application considers that the elastic wavefield contains multiple synthetic multi-component seismic data. Based on the observed seismic data, the synthetic multi-component seismic data, and the conventional L2 norm, the objective function is calculated. The formula for calculating the objective function includes:

[0086]

[0087] Describe the objective function. This represents the vertical component of the particle velocity vector in synthesized multi-component seismic data. This represents the vertical component of the particle velocity vector in synthesized multi-component seismic data. This represents the vertical component of the initial particle velocity vector in the observed seismic data. This represents the horizontal component of the initial particle velocity vector in the observed seismic data. Represents the spatial coordinates of the receiver point. This indicates the maximum sampling time.

[0088] In step S180, the weights of the preset neural network are iteratively updated using an objective function, and the error is calculated using a selected loss function. The formula for calculating the error using the loss function includes:

[0089]

[0090] In the above formula, Indicates the number of earthquake focal points. This indicates observed earthquake data. This indicates composite multi-component seismic data.

[0091] The weights of the preset neural network are iteratively updated until the error in the loss function calculation reaches a first preset threshold or the number of iterations reaches a second preset threshold. At this point, the updating of the weights of the preset neural network is paused, and the target gradient update value of the elasticity parameter is obtained. Indicatively, the first and second preset thresholds can be set according to requirements; this embodiment does not specifically limit the first and second preset thresholds.

[0092] In step S190, based on the target gradient update value and the second target elastic parameter model, a final target elastic parameter model related to the target longitudinal wave velocity, the target transverse wave velocity, and the target medium density is obtained. The second target elastic parameter model is determined based on the gradient update value and elastic parameter model obtained from the iterative update of the weights of the previous preset neural network.

[0093] This application's embodiments learn the nonlinear mapping relationship between synthesized multi-component seismic data and elastic parameter update gradients, avoiding the direct prediction of elastic parameter models as in existing technologies. This effectively alleviates the problem of local artifacts caused by the limited expressive power of neural networks and the ill-posedness problem caused by gradient crosstalk between different elastic parameters during the joint inversion of multiple elastic parameters (P-wave velocity, S-wave velocity, and medium density).

[0094] In one optional implementation, the first target elasticity parameter model includes:

[0095]

[0096] In the formula, This represents the first objective elasticity parameter model. This indicates the decoder module. Indicates encoder module, Indicates network weights, This indicates observed earthquake data. Indicates custom weights. Indicates the update step size. This represents the first initial elasticity parameter model. The custom weights are defined based on empirical formulas.

[0097] In one alternative implementation, the encoder module includes an encoder consisting of alternating 3x3 convolutional layers and 2x2 max pooling layers.

[0098] In the embodiments of this application, in the preset neural network trained by the Unet model, the encoder adopts a structure of alternating combination of 3*3 convolutional layers and 2*2 max pooling layers, which can more efficiently capture the high-dimensional feature representation of observed seismic data.

[0099] In one optional implementation, the decoder module includes a first decoder, a second decoder, and a third decoder. The first decoder is used to generate longitudinal wave velocity, the second decoder is used to generate transverse wave velocity, and the third decoder is used to generate dielectric density. The first decoder, the second decoder, and the third decoder are all composed of multiple 2*2 deconvolution layers.

[0100] In this embodiment, in the preset neural network trained by the Unet model, a multi-decoder structure is used to process the inversion of different elastic parameters. This maintains the consistency of high-dimensional feature representation extraction and enables independent decoding of gradients of different elastic parameters, effectively reducing gradient crosstalk effects between multiple elastic parameters, thereby significantly improving the resolution and robustness of the inversion process. The specific form of the first target elastic parameter model includes:

[0101]

[0102]

[0103]

[0104]

[0105] In the above formula, Indicates the first decoder, Indicates the second decoder. Indicates the third decoder. Indicates the initial longitudinal wave velocity. Indicates the initial shear wave velocity. Indicates the initial medium density. Indicates the target longitudinal wave velocity, Indicates the target shear wave velocity. Indicates the density of the target medium.

[0106] In one optional implementation, step S110 may include the following steps:

[0107] Step S111: Obtain the initial medium density, initial longitudinal wave velocity, and initial transverse wave velocity;

[0108] Step S112: Based on the initial medium density, initial longitudinal wave velocity, and initial transverse wave velocity, obtain the initial medium Lamé parameters;

[0109] Step S113: Based on the initial medium Lamé parameters and elastic wave equations, obtain the initial synthetic seismic data.

[0110] In steps S111-S112, the initial medium Lamé parameters are obtained based on the acquired initial medium density, initial P-wave velocity, initial S-wave velocity, and Lamé parameter calculation formula. The Lamé parameter calculation formula includes:

[0111]

[0112]

[0113] In the formula, and All represent the initial media Lamé parameters. Indicates the initial medium density. Indicates the initial shear wave velocity. This represents the initial longitudinal wave velocity.

[0114] In step S113, the initial medium Lame parameters are set. and Density of the medium in the elastic wave equation and Substituting these values ​​into the elastic wave equation yields the initial synthetic seismic data, which includes the horizontal component of the particle velocity vector. and the vertical plane component of the particle velocity vector The elastic wave equations include:

[0115]

[0116]

[0117]

[0118]

[0119]

[0120] In the formula, Indicates the density of the medium. Indicates the horizontal plane. Indicates a vertical plane. The horizontal component of the particle velocity vector. Indicates the propagation time of seismic waves. express Stress response in the direction, express Stress response in the direction, The vertical plane component representing the particle velocity vector. express Stress response in the direction, Indicates the media Lame parameters, This indicates the media lamellar parameters.

[0121] In one optional implementation, step S150 may include the following steps:

[0122] Step S151: Based on the P-wave velocity, S-wave velocity, and medium density, as well as the Lamé parameter calculation formula, obtain the Lamé parameters of the target medium. The Lamé parameter calculation formula includes:

[0123]

[0124]

[0125] In the formula, and All represent the Lamé parameters of the target medium. Indicates the density of the medium. Indicates the transverse wave velocity. This indicates the longitudinal wave velocity.

[0126] In step S151, the Lamé parameters of the target medium are calculated using the formulas for calculating longitudinal wave velocity, transverse wave velocity, medium density, and Lamé parameters. and .

[0127] The following embodiments of this application further verify the application effect of the elastic wave full waveform inversion method of this application through Embodiment 1 for a complex structure elastic parameter model and Embodiment 2 for an anomalous body elastic parameter model. The complex structure elastic parameter model and the anomalous body elastic parameter model are two typical problems in elastic wave full waveform inversion.

[0128] Example 1:

[0129] To verify the applicability of the elastic wave full waveform inversion method based on a single encoder-multi-decoder neural network structure proposed in this application in alleviating dependence on the initial model, Example 1 uses a two-dimensional elastic Marmousi model for numerical simulation experiments. Figure 2 This illustration schematically shows a real elastic parameter model based on a two-dimensional elastic Marmousi model simulation according to an embodiment of this application. Figure 2 (a) is the longitudinal wave velocity model simulated based on the two-dimensional elastic Marmousi model. Figure 2 (b) is the transverse wave velocity model simulated based on the two-dimensional elastic Marmousi model. Figure 2 (c) is the medium density model simulated based on the two-dimensional elastic Marmousi model. The model size of the two-dimensional elastic Marmousi model is 128×256, the spatial sampling interval is 20m, and a water layer of about 170m thickness is set in the shallow layer. The longitudinal wave velocity is... The transverse wave velocity is The density of the medium is Regarding the observation system configuration, 20 seismic sources were uniformly arranged on the upper boundary of the two-dimensional elastic Marmousi model, and 256 geophones were installed at the bottom of the water layer. The source signals used a Ricker wavelet with a dominant frequency of 10 Hz, a time sampling interval of 6 ms, and a maximum recording time of 2.4 s.

[0130] Figure 3 This illustration schematically shows a diagram of an initial elasticity parameter model according to an embodiment of this application. Figure 3 (d) represents the initial P-wave velocity model. Figure 3 (e) represents the initial shear wave velocity model. Figure 3 (f) shows the initial medium density model, and the first initial elastic parameter model is constructed using linear interpolation. A comparative experiment was conducted between the traditional physics-driven elastic wave full waveform inversion method and the elastic wave full waveform inversion method of this application. Both methods used the Adam optimization algorithm for parameter updates. The learning rate of the traditional physics-driven elastic wave full waveform inversion method was set to 5, while the learning rate of the elastic wave full waveform inversion method of this application was set to 0.001. To further alleviate the crosstalk problem in the inversion process of multiple elastic parameters, the custom weights of the elastic wave full waveform inversion method of this application are as follows: the custom weight for P-wave velocity is 2, the custom weight for S-wave velocity is 1, and the custom weight for medium density is 0.04. This custom weight design can effectively control the update amplitude of P-wave velocity, S-wave velocity, and medium density during the joint inversion process, improving the stability and resolution of parameter inversion.

[0131] In this embodiment, the Marmousi model was selected for comparative experiments. The same set of observed seismic data was input into the physical constraint-based multi-decoding neural network constructed by the traditional physical-driven elastic wave full waveform inversion method and the elastic wave full waveform inversion method of this application, respectively, and 2000 iterations were performed.

[0132] Figure 4 This illustration schematically shows a model of elastic parameters obtained by the elastic wave full waveform inversion method according to an embodiment of this application. Figure 4 As shown, Figure 4 (a) is a longitudinal wave velocity model obtained based on the elastic wave full waveform inversion method of this application. Figure 4 (b) is a transverse wave velocity model obtained based on the elastic wave full waveform inversion method of this application. Figure 4 (c) is a medium density model obtained by the elastic wave full waveform inversion method based on this application.

[0133] Figure 5 This illustration schematically shows an elastic parameter model obtained by an elastic wave full waveform inversion method based on traditional physics-driven methods according to an embodiment of this application. Figure 5 (d) shows a longitudinal wave velocity model obtained by a traditional physics-driven full-waveform inversion method for elastic waves. Figure 5 (e) shows a transverse wave velocity model obtained based on a traditional physics-driven full-waveform inversion method for elastic waves. Figure 5 (f) is a medium density model obtained by a traditional physical-driven elastic wave full waveform inversion method.

[0134] Comparison Figure 4(a)~(c) and Figure 5 (e)~(f) Traditional physics-driven full-waveform inversion methods for elastic waves are highly dependent on the initial model and prone to cycle skipping, resulting in inaccurate recovery of key model structures, particularly significant imaging deviations in shear wave velocity and medium density. In contrast, the elastic parameter model constructed based on the full-waveform inversion method of this application has a clear overall structure, with no significant interference between elastic parameters, and exhibits stronger adaptability and stability to the initial model.

[0135] To further evaluate accuracy, profiles of each elastic parameter model were extracted at the 2560th common depth point (CDP=2.56km) at the underground horizontal location. These profiles were then compared with the actual elastic parameter model and the first initial elastic parameter model. Corresponding single-channel curves for P-wave velocity, S-wave velocity, and medium density were plotted. Figure 6 A schematic diagram illustrating a comparison of single-track curves of elastic parameters according to embodiments of this application is shown, such as... Figure 6 As shown, Figure 6 (a) is a single-channel curve of the P-wave velocity. Figure 6 (b) is a single-track curve of shear wave velocity. Figure 6 (c) shows the mass density single-channel curve. It can be observed that the profile obtained by the elastic wave full waveform inversion method based on this application has a higher degree of fit with the real curve, and exhibits superior performance in characterizing elastic parameters compared to the traditional physics-driven elastic wave full waveform inversion method. Furthermore, comparing the trends of the normalized loss function of the two methods... Figure 7 This illustration schematically shows a comparison of the changing trend of a normalized loss function according to an embodiment of this application, such as... Figure 7 As shown, although the traditional physical-driven elastic wave full waveform inversion method converges quickly in the early stage, it tends to stagnate after iteration and has the risk of converging in a non-optimal direction. In contrast, the elastic wave full waveform inversion method of this application maintains stable oscillatory convergence and the overall error reduction effect is more significant.

[0136] Example 2:

[0137] To verify the applicability of the elastic wave full waveform inversion method based on a single encoder-multiple decoder neural network structure proposed in this application to the elastic parameter model of anomalies, the elastic wave full waveform inversion method of this application was applied to the two-dimensional elastic BP model, which is more difficult to invert, and numerical simulation was carried out. Figure 8 The schematic diagram illustrates a real elastic parameter model based on a two-dimensional elastic BP model simulated according to an embodiment of this application, such as... Figure 8 As shown, Figure 8 (a) is the longitudinal wave velocity model simulated based on a two-dimensional elastic BP model. Figure 8 (b) is the transverse wave velocity model simulated based on the two-dimensional elastic BP model. Figure 8(c) is the medium density model based on a two-dimensional elastic BP model simulation. The BP model has a grid size of 128×256 and a spatial sampling interval of 20m. The shallow part of the BP model contains a water layer with a thickness of approximately 170m. The elastic parameters of the BP model are: longitudinal wave velocity is... The transverse wave velocity is The density of the medium is In terms of observation system configuration, 20 seismic sources were evenly distributed at the top of the BP model, and 256 geophones were installed at the bottom of the water layer. The seismic source signals used were Ricker wavelets with a main frequency of 10 Hz, a time sampling interval of 6 ms, and a maximum recording duration of 3 s. Figure 9 This illustration schematically shows another initial elasticity parameter model according to an embodiment of this application. Figure 9 (d) represents the initial P-wave velocity model. Figure 9 (e) represents the initial shear wave velocity model. Figure 9 (f) is the initial medium density model, and the initial elastic parameter model is obtained by smoothing the real model.

[0138] This embodiment compares and analyzes the traditional full-waveform inversion method and the physics-guided multi-decoding neural network method proposed in this invention under the same optimization strategy. Both methods use the Adam algorithm for parameter iterative updates. The traditional method uses a learning rate of 5, while the method of this invention uses a smaller learning rate of 0.001 to improve network training stability. To further alleviate crosstalk issues in the inversion of multiple elastic parameters, the custom weights of the elastic wave full-waveform inversion method in this application are as follows: a custom weight of 2 for P-wave velocity, a custom weight of 1 for S-wave velocity, and a custom weight of 0.04 for medium density. This custom weight design effectively controls the update amplitude of P-wave velocity, S-wave velocity, and medium density during joint inversion, improving the stability and resolution of parameter inversion.

[0139] In this embodiment, the BP model is selected for comparative experiments. The same set of observed seismic data is input into the physical constraint-based multi-decoding neural network constructed by the traditional physical-driven elastic wave full waveform inversion method and the elastic wave full waveform inversion method of this application, respectively, and 2000 iterations are performed.

[0140] Figure 10 This illustration schematically shows another elastic parameter model obtained by the elastic wave full waveform inversion method according to an embodiment of this application, as shown below. Figure 10 As shown, Figure 10 (a) is a longitudinal wave velocity model obtained by another elastic wave full waveform inversion method based on this application. Figure 10 (b) is a transverse wave velocity model obtained by another elastic wave full waveform inversion method based on this application. Figure 10(c) is a medium density model obtained by another elastic wave full waveform inversion method based on this application.

[0141] Figure 11 This illustration schematically shows an elastic parameter model obtained by another elastic wave full waveform inversion method based on conventional physics-driven methods according to an embodiment of this application. Figure 11 (d) shows the longitudinal wave velocity model obtained by another method based on the traditional physical-driven full-waveform inversion of elastic waves. Figure 11 (e) shows a transverse wave velocity model obtained by another method based on the traditional physical-driven full-waveform inversion of elastic waves. Figure 11 (f) is a medium density model obtained by another method based on the full waveform inversion of elastic waves driven by traditional physics.

[0142] The initial elastic parameter model selected in this embodiment is relatively ideal, and the traditional physics-driven elastic wave full waveform inversion method can achieve a certain degree of structural recovery in the salt dome region. However, the traditional physics-driven elastic wave full waveform inversion method is highly dependent on the initial elastic parameter model and is prone to cycle skipping during the inversion process, especially in the reconstruction of shear wave velocity and medium density, resulting in significant errors and making it difficult to accurately identify deep structures and key geological features. In contrast, the elastic wave full waveform inversion method based on a multi-decoder neural network proposed in this application shows stronger modeling capabilities in complex subsalt structures. The inversion results obtained based on the elastic wave full waveform inversion method of this application are highly consistent with the real model, especially in the characterization of medium density and deep structures. At the same time, no obvious crosstalk occurs between multiple elastic parameters, indicating that the method has good parameter discrimination ability and stability.

[0143] Compare the trends of the normalized loss function of the two methods. Figure 12 This illustration schematically shows a comparison of the changing trend of another normalized loss function according to an embodiment of this application, such as... Figure 12 As shown, due to the relatively ideal initial elastic parameter model, both methods exhibit good convergence trends during training. However, further analysis reveals that while the traditional physics-driven elastic wave full waveform inversion method decreases rapidly in the initial stage, its convergence speed slows significantly in the later stages, posing a risk of getting trapped in local extrema and making it difficult to continuously optimize the objective function. In contrast, the elastic wave full waveform inversion method proposed in this application maintains a relatively stable oscillatory convergence characteristic throughout the entire training process, with the overall error continuously decreasing, reflecting superior global convergence capability and optimization stability.

[0144] This application also provides an elastic wave full waveform inversion system, the system comprising:

[0145] The acquisition module is used to acquire initial synthetic seismic data, observed seismic data, and a first initial elastic parameter model. The observed seismic data includes the horizontal component and the vertical component of the initial particle velocity vector.

[0146] The extraction module is used to input the observed seismic data into the encoder module of the preset neural network to extract the high-dimensional feature representation of the observed seismic data. The preset neural network is trained through the Unet model.

[0147] The first module is used to input the high-dimensional feature representation into the decoder module of the preset neural network to obtain the initial gradient update value of the elasticity parameter.

[0148] The second module is used to obtain a first target elastic parameter model related to longitudinal wave velocity, transverse wave velocity and medium density based on the initial gradient update value and the first initial elastic parameter model.

[0149] The third module is used to obtain the Lamé parameters of the target medium based on the longitudinal wave velocity, transverse wave velocity, and medium density.

[0150] The fourth module is used to obtain synthetic multi-component seismic data based on the Lamé parameters of the target medium and the elastic wave equation. The synthetic multi-component seismic data includes the horizontal component and the vertical component of the target particle velocity vector.

[0151] The fifth module is used to obtain the objective function based on observed seismic data and synthetic multi-component seismic data;

[0152] The iterative update module is used to iteratively update the weights of the preset neural network using the objective function until the error obtained based on the observed seismic data, the synthesized multi-component seismic data and the loss function reaches the first preset threshold or the number of iterations reaches the second preset threshold, so as to obtain the target gradient update value of the elastic parameter.

[0153] The sixth module is used to obtain the final target elastic parameter model related to the target longitudinal wave velocity, the target transverse wave velocity, and the target medium density based on the target gradient update value and the second target elastic parameter model. The second target elastic parameter model is determined based on the gradient update value and elastic parameter model obtained by the iterative update of the weights of the previous preset neural network.

[0154] It is understood that the elastic wave full waveform inversion system provided in this application embodiment can realize each process of the elastic wave full waveform inversion method in the above embodiment and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0155] This application also provides an elastic wave full waveform inversion device, including:

[0156] The memory is configured to store instructions;

[0157] The processor is configured to retrieve instructions from memory and, when executing instructions, to implement the elastic wave full waveform inversion method as described above.

[0158] It is understood that the elastic wave full waveform inversion device provided in this application embodiment can realize each process of the elastic wave full waveform inversion method in the above embodiment and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0159] This application also provides a machine-readable storage medium storing instructions for causing a machine to execute the elastic wave full waveform inversion method as described above.

[0160] It is understood that the machine-readable storage medium provided in the embodiments of this application can implement each process of the elastic wave full waveform inversion method in the above embodiments and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0161] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0162] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0163] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0164] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0165] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0166] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0167] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0168] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0169] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for full waveform inversion of elastic waves, characterized in that, The method includes: Acquire initial synthetic seismic data, observed seismic data, and a first initial elastic parameter model, wherein the observed seismic data includes the horizontal component and the vertical component of the initial particle velocity vector; The observed seismic data is input into the encoder module of a preset neural network to extract high-dimensional feature representations of the observed seismic data, wherein the preset neural network is trained using the Unet model; The high-dimensional feature representation is input into the decoder module of the preset neural network to obtain the initial gradient update value of the elasticity parameter; Based on the initial gradient update value and the first initial elastic parameter model, a first target elastic parameter model related to the longitudinal wave velocity, transverse wave velocity and medium density is obtained. Based on the longitudinal wave velocity, the transverse wave velocity, and the medium density, the Lamé parameters of the target medium are obtained; Based on the Lamé parameters of the target medium and the elastic wave equation, synthetic multi-component seismic data is obtained, wherein the synthetic multi-component seismic data includes the horizontal component and the vertical component of the target particle velocity vector. Based on the observed seismic data and the synthesized multi-component seismic data, the objective function is obtained; The weights of the preset neural network are iteratively updated using the objective function until the error obtained based on the observed seismic data, the synthetic multi-component seismic data, and the loss function reaches a first preset threshold or the number of iterations reaches a second preset threshold, thereby obtaining the target gradient update value of the elastic parameter. Based on the target gradient update value and the second target elastic parameter model, a final target elastic parameter model related to the target longitudinal wave velocity, the target transverse wave velocity, and the target medium density is obtained. The second target elastic parameter model is determined based on the gradient update value and elastic parameter model obtained from the previous iteration update of the weights of the preset neural network.

2. The method according to claim 1, characterized in that, The first target elasticity parameter model includes: In the formula, This represents the elasticity parameter model of the first target. This refers to the decoder module. This refers to the encoder module. Indicates network weights, This refers to the observed seismic data. Indicates custom weights. Indicates the update step size. This represents the first initial elastic parameter model.

3. The method according to claim 1, characterized in that, The process of obtaining the target medium Lamé parameters based on the longitudinal wave velocity, the transverse wave velocity, and the medium density includes: Based on the longitudinal wave velocity, the transverse wave velocity, and the medium density, as well as the Lamé parameter calculation formula, the Lamé parameters of the target medium are obtained, wherein the Lamé parameter calculation formula includes: In the formula, and All of these represent the Lamé parameters of the target medium. This indicates the density of the medium. This indicates the transverse wave velocity. This indicates the longitudinal wave velocity.

4. The method according to claim 1, characterized in that, The acquisition of initial synthetic seismic data includes: Obtain the initial medium density, initial longitudinal wave velocity, and initial transverse wave velocity; Based on the initial medium density, the initial longitudinal wave velocity, and the initial transverse wave velocity, the initial medium Lamé parameters are obtained; The initial synthetic seismic data are obtained based on the initial medium Lamé parameters and the elastic wave equation.

5. The method according to claim 4, characterized in that, The elastic wave equation includes: In the formula, Indicates the density of the medium. Indicates the horizontal plane. Indicates a vertical plane. The horizontal component of the particle velocity vector. Indicates the propagation time of seismic waves. express Stress response in the direction, express Stress response in the direction, The vertical plane component representing the particle velocity vector. express Stress response in the direction, Indicates the media Lame parameters, This indicates the media lamellar parameters.

6. The method according to claim 1, characterized in that, The encoder module includes an encoder, which is composed of multiple alternating 3*3 convolutional layers and 2*2 max pooling layers.

7. The method according to claim 1, characterized in that, The decoder module includes a first decoder, a second decoder, and a third decoder. The first decoder is used to generate the longitudinal wave velocity, the second decoder is used to generate the transverse wave velocity, and the third decoder is used to generate the medium density. The first decoder, the second decoder, and the third decoder are all composed of multiple 2*2 deconvolution layers.

8. A full waveform inversion system for elastic waves, characterized in that, The system includes: The acquisition module is used to acquire initial synthetic seismic data, observed seismic data, and a first initial elastic parameter model, wherein the observed seismic data includes the horizontal component and the vertical component of the initial particle velocity vector; An extraction module is used to input the observed seismic data into an encoder module of a preset neural network to extract high-dimensional feature representations of the observed seismic data, wherein the preset neural network is trained using a Unet model; The first obtaining module is used to input the high-dimensional feature representation into the decoder module of the preset neural network to obtain the initial gradient update value of the elasticity parameter; The second obtaining module is used to obtain a first target elastic parameter model related to longitudinal wave velocity, transverse wave velocity and medium density based on the initial gradient update value and the first initial elastic parameter model. The third module is used to obtain the Lamé parameters of the target medium based on the longitudinal wave velocity, the transverse wave velocity, and the medium density. The fourth module is used to obtain synthetic multi-component seismic data based on the Lamé parameters of the target medium and the elastic wave equation, wherein the synthetic multi-component seismic data includes the horizontal component and the vertical component of the target particle velocity vector. The fifth module is used to obtain the objective function based on the observed seismic data and the synthetic multi-component seismic data; An iterative update module is used to iteratively update the weights of the preset neural network using the objective function until the error obtained based on the observed seismic data, the synthetic multi-component seismic data, and the loss function reaches a first preset threshold or the number of iterative updates reaches a second preset threshold, thereby obtaining the target gradient update value of the elastic parameter. The sixth module is used to obtain a final target elastic parameter model related to the target longitudinal wave velocity, the target transverse wave velocity, and the target medium density based on the target gradient update value and the second target elastic parameter model. The second target elastic parameter model is determined based on the gradient update value and elastic parameter model obtained from the previous iteration update of the weights of the preset neural network.

9. A device for inverting the full waveform of an elastic wave, characterized in that, include: The memory is configured to store instructions; The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the elastic wave full waveform inversion method according to any one of claims 1 to 7.

10. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to perform the elastic wave full waveform inversion method according to any one of claims 1 to 7.

Images