Seismic full waveform inversion method and device, electronic equipment and storage medium

Abstract

This application provides a method, apparatus, electronic device, and storage medium for seismic full waveform inversion, comprising: inputting random variables into a pre-constructed convolutional neural network model, outputting parameter variable data through the convolutional neural network model, and determining physical property parameter data based on the parameter variable data and pre-acquired initial parameter data; performing forward modeling on the physical property parameter data to obtain simulated seismic data; calculating the residual value between the simulated seismic data and pre-acquired actual seismic data, and, in response to the residual value being less than or equal to a preset threshold, using the physical property parameter data as the target physical property parameter data for the seismic full waveform; the seismic full waveform inversion method of this invention employs convolutional neural network reparameterization, which can effectively weaken the influence of cycle skipping and effectively improve the convergence ability and robustness of elastic full waveform inversion.

Description

Technical Field

[0001] This application relates to the field of oil and gas exploration and development technology, and in particular to a seismic full waveform inversion method, apparatus, electronic equipment and storage medium. Background Technology

[0002] Elastic full-waveform inversion is a data-driven, high-resolution seismic data processing method that can reveal information such as P-wave and S-wave velocities and densities in subsurface strata. This method utilizes the kinematic and dynamic information in pre-stack shot gathering seismic data to minimize the residuals between simulated and actual data, thereby inverting complex subsurface structures and rock types.

[0003] Currently, the traditional elastic full waveform inversion method uses the initial model as the optimization parameter and directly updates the model. The disadvantage of this method is that the simultaneous updating of multiple parameters is affected by coupling. An erroneous update in one parameter will affect other parameters, which will cause errors to accumulate. The inversion is more likely to get trapped in local minima, causing problems such as "cycle skipping" and insufficient convergence. Summary of the Invention

[0004] In view of this, the purpose of this application is to propose a method, apparatus, electronic equipment and storage medium for seismic full waveform inversion, so as to solve or partially solve the above problems.

[0005] To achieve the above objectives, this application provides a seismic full-waveform inversion method, the method comprising:

[0006] Random variables are input into a pre-built convolutional neural network model, and parameter variable data are output through the convolutional neural network model. Based on the parameter variable data and the pre-acquired initial parameter data, physical property parameter data is determined.

[0007] The physical property parameter data are subjected to forward modeling to obtain simulated seismic data;

[0008] The residual value between the simulated earthquake data and the pre-acquired actual earthquake data is calculated. In response to the residual value being less than or equal to a preset threshold, the physical property parameter data is used as the target physical property parameter data for the full earthquake waveform.

[0009] Based on the same inventive concept, this application also provides a seismic full-waveform inversion device, the device comprising:

[0010] The data acquisition module is used to input random variables into a pre-built convolutional neural network model, output parameter variable data through the convolutional neural network model, and determine physical property parameter data based on the parameter variable data and the pre-acquired initial parameter data.

[0011] The full waveform forward modeling module is used to perform forward modeling on the physical property parameter data to obtain simulated seismic data;

[0012] The full waveform inversion module is used to calculate the residual value between the simulated seismic data and the pre-acquired actual seismic data. In response to the residual value being less than or equal to a preset threshold, the physical property parameter data is used as the target physical property parameter data of the seismic full waveform.

[0013] Based on the same inventive concept, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements a seismic full waveform inversion method as described above when executing the computer program.

[0014] Based on the same inventive concept, this application also provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute a seismic full-waveform inversion method as described above.

[0015] As can be seen from the above, the seismic full waveform inversion method, apparatus, electronic device, and storage medium provided in this application construct a convolutional neural network model, determine physical property parameter data based on parameter variable data and initial parameter data, perform forward modeling on the physical property parameter data to obtain simulated seismic data, calculate the residual value between the simulated seismic data and the pre-acquired actual seismic data, determine whether the residual value meets the preset conditions, and then determine the target physical property parameter data of the seismic full waveform. This application uses a convolutional neural network to reconstruct the gradient of the loss function on the elastic parameters, which to a certain extent weakens the influence of the coupling effect between elastic parameters, reduces the propagation of errors between different models, effectively avoids cycle jumps, and improves the full convergence capability of seismic full waveform inversion. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of the seismic full-waveform inversion method according to an embodiment of this application;

[0018] Figure 2 This is a flowchart illustrating another embodiment of the seismic full-waveform inversion method of this application;

[0019] Figure 3These are comparison charts showing the testing of various models in the embodiments of this application;

[0020] Figure 4 This is a schematic diagram of the structure of the seismic full-waveform inversion device according to an embodiment of this application;

[0021] Figure 5 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0023] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0024] Currently, existing elastic full-waveform inversion methods primarily update the initial elastic model. Based on the least-squares best-fit principle, a suitable objective function is constructed. Then, the gradient of the objective function with respect to the initial elastic model is calculated using the adjoint state method (i.e., the cross-correlation between the forward wavefield and the adjoint wavefield). The update method for the initial model is determined based on this gradient. In practical applications, since inversion is a highly nonlinear process, conventional methods can easily lead to the "cycle skipping" problem, causing the inversion to get trapped in local minima. Furthermore, elastic full-waveform inversion requires updating multiple parameters simultaneously, and the loss function has a coupling effect on the gradients of these parameters. This effect exacerbates the nonlinearity problem of the inversion. Therefore, improving the robustness of the inversion is of great value.

[0025] Therefore, the present invention provides a method, apparatus, electronic device and storage medium for seismic full waveform inversion.

[0026] This invention replaces the initial elasticity model update with a convolutional neural network model. Based on the convolutional kernel inherent in the convolutional neural network model, it increases the spatial correlation in the model as a regularization strategy to improve the robustness of the inversion, enabling the inversion to escape local minima. Even under abrupt changes in velocity, the inversion can still maintain convergence.

[0027] Meanwhile, the convolutional neural network reconstructs the gradient of the loss function with respect to the elasticity parameters (including the first parameter λ, the second parameter μ, and the density ρ), which to some extent weakens the influence of the coupling between elasticity parameters, alleviates the propagation of errors between different models, and thus improves the convergence ability of elastic full waveform inversion.

[0028] See appendix Figure 1 and attached Figure 2 , attached Figure 1 This is a schematic flowchart of a seismic full-waveform inversion method according to an embodiment of the present invention. Figure 2 This is a schematic flowchart of a seismic full-waveform inversion method according to another embodiment of the present invention; the following will be illustrated in conjunction with the attached diagram. Figure 1 and attached Figure 2 The earthquake full waveform inversion method proposed in this application is described in detail.

[0029] like Figure 1 As shown, an earthquake full waveform inversion method in this embodiment of the invention mainly includes the following steps S1-S3.

[0030] Step S1: Input random variables into a pre-built convolutional neural network model, and output parameter variable data through the convolutional neural network model. The parameter variable data includes, but is not limited to, the change in P-wave velocity (i.e., ΔP-wave velocity) and the change in S-wave velocity (i.e., ΔS-wave velocity). Based on the parameter variable data and the pre-acquired initial parameter data, determine the physical property parameter data. The initial parameter data includes, but is not limited to, the initial P-wave velocity and the initial S-wave velocity. The physical property parameter data includes, but is not limited to, the P-wave velocity, the S-wave velocity, and the density.

[0031] In some embodiments, x, the input to the convolutional neural network model, is treated as a random variable. After processing by the convolutional neural network model, the changes in longitudinal wave velocity (i.e., ΔL-wave velocity) and transverse wave velocity (i.e., ΔS-wave velocity) are output. The pre-acquired initial longitudinal wave velocity and initial transverse wave velocity are added to the changes in longitudinal wave velocity and transverse wave velocity output by the convolutional neural network, respectively, to obtain the longitudinal wave velocity and transverse wave velocity.

[0032] This technical solution uses the following neural network reparameterization formula to determine the P-wave velocity and S-wave velocity, namely:

[0033] V p =CNN vp(x, w) vp )+V pinit

[0034] V s =CNN vs (x, w) vs )+V sinit

[0035] In the formula, x represents a random variable, CNN vp and CNN vs These are the convolutional neural network models corresponding to the longitudinal wave velocity and the transverse wave velocity, respectively. vp and w vs CNN vp and CNN vs The parameters to be optimized in V (the parameters to be optimized refer to the parameters in the convolutional neural network model, such as the weights and biases of the convolutional layers) pinit V is expressed as the initial longitudinal wave velocity. sinit V is represented as the initial transverse wave velocity. p Expressed as longitudinal wave velocity, V s It is expressed as transverse wave velocity.

[0036] It should be noted that the convolutional neural network used in this technical solution is a conventionally constructed convolutional neural network by those skilled in the art, which can be implemented using existing technology in this field. The convolutional neural network can process the input random variable x and output Δ longitudinal wave velocity and Δ transverse wave velocity. The structure of the convolutional neural network model will not be elaborated on here. Based on the convolution kernel in the convolutional neural network model, the spatial correlation in the model is increased to improve the robustness of the inversion, making the inversion less likely to get trapped in local minima and ensuring full convergence of the inversion.

[0037] Step S2: The physical property parameter data is forward modeled to obtain simulated seismic data.

[0038] Specifically, firstly, the Lamé parameters are calculated based on the shear wave velocity and P-wave velocity obtained in step S1 using the Lamé parameter formula. Then, the density ρ and the calculated Lamé parameters are input into the elastic wave equation for forward modeling to generate simulated seismic data, which is obtained by sampling the forward modeled wave field.

[0039] The formula for the Lamé parameter is:

[0040] λ=(V p 2 -2V s 2 )ρ

[0041] μ = V s 2 ρ

[0042] Where ρ represents density, λ represents the first parameter, and μ represents the second parameter;

[0043] It is important to explain here that in continuum mechanics, the Lamé parameter (also known as the Lamé constant) is two material-related quantities represented by λ and μ, which appear in the strain-stress relationship; usually, λ and μ are referred to as the first parameter and the second parameter of the Lamé parameter, respectively.

[0044] The forward wave field is determined using the elastic wave equation based on the Lamé parameters. The forward wave field includes a forward transverse normal stress wave field and a forward longitudinal normal stress wave field. The elastic wave equation is as follows:

[0045]

[0046] In the formula, v x Represented as a forward-modeled transverse velocity wave field, v z Represented as a forward-modeled longitudinal velocity wave field, the difference from the velocity model described above is that v here... x v z It represents the transverse and longitudinal components of the seismic wave amplitude, σ xx Represented as a forward-modeled transverse normal stress wave field, σ zz Represented as the forward-modeled longitudinal normal stress wave field, σ xz Let λ represent the forward-modeled shear stress wave field, λ represent the first parameter, μ represent the second parameter, and ρ represent the density. It is expressed as the derivative of the wave field with respect to the transverse direction. Let f be the derivative of the wave field with respect to the longitudinal direction. x Represented as a transverse normal stress source, f z Represented as a longitudinal normal stress source;

[0047] Based on the forward-modeled transverse normal stress wave field σ xx and the forward-modeled longitudinal normal stress wave field σ zz The simulated earthquake data d is obtained, and the formula for the simulated earthquake data d is:

[0048]

[0049] Step S3: Calculate the residual value between the simulated earthquake data and the pre-acquired actual earthquake data. In response to the residual value being less than or equal to a preset threshold, use the physical property parameter data as the target physical property parameter data for the full earthquake waveform.

[0050] Specifically, by calculating the residual value between simulated earthquake data and actual earthquake data, it is determined whether the residual value meets the preset conditions. If the residual value is less than or equal to the preset threshold, the convolutional neural network model does not need to be updated, and the determined physical property parameter data is used as the target physical property parameter data. If the residual value is greater than the preset threshold, the convolutional neural network model is updated. The updated physical property parameter data is determined by the updated parameter variable data output by the updated convolutional neural network model and the pre-acquired initial parameter data, and the updated physical property parameter data is used as the target physical property parameter data.

[0051] In some embodiments, the method further includes: responding to the residual value being greater than a preset threshold, constructing a loss function based on the residual value, calculating the gradient of the loss function with respect to the convolutional neural network model, updating the parameters of the convolutional neural network model based on the gradient, outputting updated parameter variable data through the updated convolutional neural network model, determining updated physical property parameter data based on the updated parameter variable data and pre-acquired initial parameter data, and using the updated physical property parameter data as the target physical property parameter data.

[0052] In some embodiments, constructing a loss function based on the residual values ​​includes:

[0053] Based on the simulated earthquake data d and the actual earthquake data d obs The loss function J is constructed from the residual values ​​between the two values, and the formula for the loss function J is as follows:

[0054]

[0055] N s and N t These represent the number of cannons and the time step per cannon, respectively. and d i，t These are the actual and simulated seismic record values ​​at time t, generated by the i-th shot, respectively.

[0056] It should be noted that in practical applications, actual earthquake data refers to the earthquake data received by the receiver. In model testing, earthquake data from the actual model is usually used instead of actual earthquake data.

[0057] In some embodiments, the gradient calculation, based on the calculation of the gradient of the loss function with respect to the convolutional neural network model, includes:

[0058] The loss function is defined as the objective function. The cross-correlation between the forward modeled seismic wavefield (including the forward modeled transverse normal stress wavefield and the forward modeled longitudinal normal stress wavefield) and the adjoint seismic wavefield (including the adjoint longitudinal normal stress wavefield and the adjoint transverse normal stress wavefield) is used to calculate the derivative of the objective function with respect to the first parameter, the second parameter, and the density using the adjoint state algorithm.

[0059] Based on the density, the P-wave velocity, the S-wave velocity, and the derivatives of the loss function with respect to the first and second parameters, the derivatives of the loss function with respect to the P-wave velocity and the S-wave velocity are calculated using the chain rule.

[0060] The gradient of the loss function with respect to the parameters in the convolutional neural network model is determined by calculating the derivatives of the loss function with respect to the P-wave velocity and the S-wave velocity using an automatic differentiation algorithm.

[0061] The formula for the adjoint state algorithm is as follows:

[0062]

[0063]

[0064]

[0065] In the formula, and Let these be the accompanying longitudinal normal stress wave field and the accompanying transverse normal stress wave field at time t+1, respectively. Represented as the accompanying shear stress wave field, This is represented as the forward-modeled transverse normal stress wave field at time t. Let be the forward-modeled longitudinal normal stress wave field at time t. Let N be the associated shear stress wave field at time t. t The total time length is represented by Δt, and the time interval is represented by Δt. and These are respectively represented as the derivatives of the objective function with respect to the elastic parameters (λ, μ, ρ).

[0066] Specifically, the loss function is defined as the objective function, and then the derivatives of the objective function with respect to the P-wave velocity and S-wave velocity are calculated using the adjoint state algorithm, such as... Figure 2 As shown by the dashed arrow, the adjoint state algorithm uses the cross-correlation between the forward-modeled seismic wavefield and the adjoint seismic wavefield to calculate the derivative, which is an effective gradient calculation method.

[0067] The chain rule calculation formula is as follows:

[0068]

[0069]

[0070] In the formula, The objective function is given by the longitudinal wave velocity V. p The derivative, The objective function is given by the shear wave velocity V. s The derivative;

[0071] In traditional methods, the initial model is updated using the aforementioned transformation formula. However, in this invention, the value to be updated becomes the elasticity parameter in the convolutional neural network model, which is then updated using an automatic differentiation algorithm. Differential calculations are performed to determine the gradient of the objective function with respect to the parameters in the convolutional neural network. The automatic differentiation algorithm formula is as follows:

[0072]

[0073] Let w be the derivative of the loss function with respect to the parameter w in the neural network model. For the parameter w to be optimized vp The derivative, For the parameter w to be optimized vs The derivative of .

[0074] Automatic differentiation is a commonly used technique in neural network training. Based on the chain rule, it greatly simplifies the complexity of gradient calculation and eliminates the need for manual calculation. When calculating complex gradients, automatic differentiation does not accumulate errors, thus solving the problem that traditional methods are prone to getting trapped in local minima during the inversion process, causing "cycle jumps" and insufficient convergence.

[0075] based on Figure 2 The schematic diagram of the seismic full waveform inversion method shown can be understood as follows:

[0076] Initially, a random variable x is input, which is processed by a convolutional neural network model to output ΔP-wave velocity and ΔS-wave velocity. These are then added to the pre-acquired initial P-wave velocity and initial S-wave velocity to obtain the P-wave velocity and S-wave velocity. The seismic wave field is obtained by forward modeling using the elastic wave equation, and simulated seismic data is obtained by sampling. The residual value between the simulated seismic data and the pre-acquired actual seismic data is calculated, and the residual value between the simulated seismic data and the actual seismic data obtained by forward modeling is compared with a preset threshold.

[0077] In response to the residual value being less than or equal to a preset threshold, the P-wave velocity and the S-wave velocity are taken as the target P-wave velocity and target S-wave velocity of the entire seismic waveform.

[0078] In response to the residual value being greater than a preset threshold, a loss function is constructed based on the residual value, and the loss function is defined as the objective function. The gradient of the objective function with respect to the P-wave velocity and S-wave velocity is calculated using the adjoint state algorithm. Then, the gradient of the objective function with respect to the parameters of the convolutional neural network model is calculated using the chain rule and automatic differentiation method. The parameters of the convolutional neural network model are updated based on the gradient calculation. The forward modeling is iterated and updated repeatedly until the residual value between the simulated seismic data and the actual seismic data obtained from the forward modeling is less than or equal to the preset threshold. Then, the update iteration is stopped. The P-wave velocity and S-wave velocity variables output by the convolutional neural network model after the iteration are calculated and the initial P-wave velocity and initial S-wave velocity are used to determine the updated P-wave velocity and S-wave velocity. The updated P-wave velocity and S-wave velocity are used as the target P-wave velocity and target S-wave velocity.

[0079] This invention employs a convolutional neural network to reparameterize the elastic model in full-waveform inversion, establishing a new full-waveform inversion process. The output of the convolutional neural network replaces the traditional initial model. A combination of automatic differentiation and adjoint state methods is used to calculate the gradient of the loss function with respect to the parameters in the convolutional neural network model. Based on the gradient, the convolutional neural network model is updated iteratively to obtain the target P-wave velocity and target S-wave velocity of the full seismic waveform. Using this method for elastic full-waveform inversion can effectively reduce the impact of "cycle skipping" and ensure sufficient convergence of the inversion. Even in models with abrupt velocity changes, this method can achieve good inversion results.

[0080] In a specific embodiment, taking the 2004BP salt body model for model testing as an example, the model size is 10100m×3400m, the model discrete interval is 50m, the source of each shot uses the Ricker wavelet with a dominant frequency of 10Hz, the number of shots is 48, the number of receivers is 202, and the total duration is 2s. Since the model contains a large number of high-speed salt bodies, it will produce large velocity abrupt changes, which poses a great challenge to traditional inversion methods.

[0081] Figure 3 The figure shows the 2004BP salt body model and inversion results. The horizontal axis represents the horizontal distance of the model, and the vertical axis represents the depth.

[0082] Figure 3In the figures, (a) and (c) represent the P-wave and S-wave velocities of the actual model, respectively; (b) and (d) represent the P-wave and S-wave velocities of the initial model, respectively; (e) and (h) represent the P-wave and S-wave velocities obtained by inversion using this method; and (f) and (g) represent the P-wave and S-wave velocities obtained by inversion using the traditional method. It is evident from the figures that for models with abrupt velocity changes, the traditional method is severely affected by cycle skipping and cannot invert the structure and velocity magnitude of the salt body. The method proposed in this invention significantly improves upon this deficiency; even in cases of abrupt velocity changes, this method can still invert the position and velocity magnitude of the salt body.

[0083] Compared to conventional methods, the convolutional neural network-based parametric elastic full-waveform inversion method proposed in this paper offers greater flexibility. The convolutional kernels in the convolutional neural network enhance the spatial correlation of the model, acting as a regularization strategy to prevent the inversion from getting trapped in local minima. This method can significantly improve inversion performance in complex salt body models, where existing techniques often fail to converge adequately due to the numerous velocity abrupt changes in the salt body model.

[0084] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0085] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0086] Based on the same inventive concept, corresponding to any of the methods in the above embodiments, this application also provides a seismic full-waveform inversion device. Please refer to [link to relevant documentation]. Figure 4 The main structural block diagram of the seismic full-waveform inversion device shown is explained below, such as... Figure 4 As shown, the seismic full waveform inversion device in this embodiment of the invention mainly includes: a data acquisition module 11, a full waveform forward modeling module 12, and a full waveform inversion module 13.

[0087] In some embodiments, the data acquisition module 11 is configured to input random variables into a pre-constructed convolutional neural network model, output parameter variable data through the convolutional neural network model, and determine physical property parameter data based on the parameter variable data and pre-acquired initial parameter data. The full waveform forward modeling module 12 is configured to perform forward modeling processing on the physical property parameter data to obtain simulated seismic data. The full waveform inversion module 13 is configured to calculate the residual value between the simulated seismic data and the pre-acquired actual seismic data, and in response to the residual value being less than or equal to a preset threshold, use the physical property parameter data as the target physical property parameter data for the full seismic waveform.

[0088] In some embodiments, the initial parameter data includes at least the initial P-wave velocity and the initial S-wave velocity, and the physical property parameter data includes at least the P-wave velocity and the S-wave velocity. The data acquisition module 11 is further configured to determine the physical property parameter data according to the following formula:

[0089] V p =CNN vp (x, w) vp )+V pinit

[0090] V s =CNN vs (x, w) vs )+V sinit

[0091] In the formula, x represents a random variable, CNN vp and CNN vs These are the convolutional neural network models corresponding to the longitudinal wave velocity and the transverse wave velocity, respectively. vp and w vs CNN vp and CNN vs The parameter to be optimized in V pinit V is expressed as the initial longitudinal wave velocity. sinit V is represented as the initial transverse wave velocity. p Represented as P-wave velocity data, V s It is represented as shear wave velocity data.

[0092] In some embodiments, the full waveform forward modeling module 12 is further configured to:

[0093] Lamé parameters are calculated based on the P-wave velocity and the S-wave velocity. The Lamé parameters include a first parameter and a second parameter, and the calculation formulas for the first parameter and the second parameter are as follows:

[0094] λ=(V p 2 -2V s 2 )ρ

[0095] μ = V s 2 ρ

[0096] Where ρ represents density, λ represents the first parameter, and μ represents the second parameter;

[0097] The forward wave field is determined using the elastic wave equation based on the Lamé parameters. The forward wave field includes a forward transverse normal stress wave field and a forward longitudinal normal stress wave field. The elastic wave equation is as follows:

[0098]

[0099] In the formula, v x Represented as a forward-modeled transverse velocity wave field, v z Represented as the forward longitudinal velocity wave field, σ xx Represented as a forward-modeled transverse normal stress wave field, σ zz Represented as the forward-modeled longitudinal normal stress wave field, σ xz Let λ represent the forward-modeled shear stress wave field, λ represent the first parameter, μ represent the second parameter, and ρ represent the density. It is expressed as the derivative of the wave field with respect to the transverse direction. Let f be the derivative of the wave field with respect to the longitudinal direction. x Represented as a transverse normal stress source, f z Represented as a longitudinal normal stress source;

[0100] Based on the forward-modeled transverse normal stress wave field σ xx and the forward-modeled longitudinal normal stress wave field σ zz The simulated earthquake data d is obtained, and the formula for the simulated earthquake data d is:

[0101]

[0102] In some embodiments, the full waveform inversion module 13 is further configured to:

[0103] In response to the residual value being greater than a preset threshold, a loss function is constructed based on the residual value, the gradient of the loss function with respect to the convolutional neural network model is calculated, the parameters of the convolutional neural network model are updated based on the gradient, updated parameter variable data are output through the updated convolutional neural network model, updated physical property parameter data are determined based on the updated parameter variable data and the pre-acquired initial parameter data, and the updated physical property parameter data is used as the target physical property parameter data.

[0104] In some embodiments, the full waveform inversion module 13 is further configured to:

[0105] Based on the simulated earthquake data d and the actual earthquake data dobs The loss function J is constructed from the residual values ​​between the two values, and the formula for the loss function J is as follows:

[0106]

[0107] N s and N t These represent the number of cannons and the time step per cannon, respectively. and d i，t These are the actual seismic data and simulated seismic data generated at time t, respectively, for the i-th shot.

[0108] In some embodiments, the full waveform inversion module 13 is further configured to:

[0109] The derivatives of the loss function with respect to the first parameter, the second parameter, and the density are calculated using the adjoint state algorithm. Based on the density, the P-wave velocity, the S-wave velocity, and the derivatives of the loss function with respect to the first parameter and the second parameter, the derivatives of the loss function with respect to the P-wave velocity and the S-wave velocity are calculated using the chain rule.

[0110] The gradient of the loss function with respect to the parameters in the convolutional neural network model is determined by calculating the derivatives of the loss function with respect to the P-wave velocity and the S-wave velocity using an automatic differentiation algorithm.

[0111] The formula for the adjoint state algorithm is as follows:

[0112]

[0113]

[0114]

[0115] In the formula, and Let these be the accompanying longitudinal normal stress wave field and the accompanying transverse normal stress wave field at time t+1, respectively. Represented as the accompanying shear stress wave field, This is represented as the forward-modeled transverse normal stress wave field at time t. Let be the forward-modeled longitudinal normal stress wave field at time t. Let N be the associated shear stress wave field at time t. t The total time length is represented by Δt, and the time interval is represented by Δt. and These are respectively expressed as the derivatives of the loss function with respect to the first parameter λ, the second parameter μ, and the density ρ.

[0116] The chain rule calculation formula is as follows:

[0117]

[0118]

[0119] In the formula, Let be the derivative of the loss function with respect to the P-wave velocity. This is the derivative of the loss function with respect to the shear wave velocity;

[0120] The automatic differentiation algorithm formula is as follows:

[0121]

[0122] Let w be the derivative of the loss function with respect to the parameter w in the neural network model. For the parameter w to be optimized vp The derivative, For the parameter w to be optimized vs The derivative of .

[0123] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0124] The apparatus described above is used to implement the corresponding seismic full waveform inversion method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0125] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the seismic full waveform inversion method described in any of the above embodiments.

[0126] Figure 5 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0127] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0128] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0129] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0130] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0131] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0132] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0133] The electronic devices described above are used to implement the corresponding seismic full-waveform inversion method in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0134] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the seismic full-waveform inversion method as described in any of the above embodiments.

[0135] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, 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, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0136] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the seismic full waveform inversion method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0137] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0138] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0139] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0140] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A seismic full-waveform inversion method, characterized in that, The method includes: A random vector is input into a pre-constructed convolutional neural network model, which outputs parameter variable data. Based on the parameter variable data and pre-acquired initial parameter data, physical property parameter data is determined. The parameter variable data includes P-wave velocity perturbation and S-wave velocity perturbation, and the initial parameter data includes initial P-wave velocity and initial S-wave velocity. The physical property parameter data are subjected to forward modeling to obtain simulated seismic data; The residual value between the simulated earthquake data and the pre-acquired actual earthquake data is calculated. In response to the residual value being less than or equal to a preset threshold, the physical property parameter data is used as the target physical property parameter data for the full earthquake waveform.

2. The seismic full-waveform inversion method according to claim 1, characterized in that, The physical property parameter data includes at least P-wave velocity and S-wave velocity, and the determination of physical property parameter data based on the parameter variable data and pre-acquired initial parameter data includes: Determine the physical property parameters using the following formula: ； ； In the formula, x represents a random vector. and These are the convolutional neural network models corresponding to the longitudinal wave velocity and the transverse wave velocity, respectively. and They are respectively and The parameters to be optimized in Represented as the initial P-wave velocity, Represented as the initial shear wave velocity, Expressed as longitudinal wave velocity, It is expressed as transverse wave velocity.

3. The seismic full-waveform inversion method according to claim 2, characterized in that, The process of performing forward modeling on the physical property parameter data to obtain simulated seismic data includes: Lamé parameters are calculated based on the P-wave velocity and the S-wave velocity. The Lamé parameters include a first parameter and a second parameter, and the calculation formulas for the first parameter and the second parameter are as follows: ； ； in, Represented as density, Represented as the first parameter, This is represented as the second parameter; The forward wave field is determined using the elastic wave equation based on the Lamé parameters. The forward wave field includes a forward transverse normal stress wave field and a forward longitudinal normal stress wave field. The elastic wave equation is as follows: ； In the formula, Represented as a forward-modeled transverse velocity wave field, Represented as a forward-modeled longitudinal velocity wave field, Represented as a forward-modeled transverse normal stress wave field, Represented as a forward-modeled longitudinal normal stress wave field, Represented as a forward-modeled shear stress wave field, Represented as the first parameter, Represented as the second parameter, Represented as density, It is expressed as the derivative of the wave field with respect to the transverse direction. It is expressed as the derivative of the wave field with respect to the longitudinal direction. This is represented as a transverse normal stress source. Represented as a longitudinal normal stress source; Based on the forward-modeled transverse normal stress wave field and the forward-modeled longitudinal normal stress wave field Obtain simulated earthquake data The simulated earthquake data The formula is: 。 4. The seismic full-waveform inversion method according to claim 3, characterized in that, Also includes: In response to the residual value being greater than a preset threshold, a loss function is constructed based on the residual value, the gradient of the loss function with respect to the convolutional neural network model is calculated, the parameters of the convolutional neural network model are updated based on the gradient, updated parameter variable data are output through the updated convolutional neural network model, updated physical property parameter data are determined based on the updated parameter variable data and the pre-acquired initial parameter data, and the updated physical property parameter data is used as the target physical property parameter data.

5. The seismic full-waveform inversion method according to claim 4, characterized in that, The construction of the loss function based on the residual value includes: Based on the simulated earthquake data Compared with actual earthquake data The loss function J is constructed from the residual values ​​between the two values, and the formula for the loss function J is as follows: ； and These represent the number of cannons and the time step per cannon, respectively. and The first Produced by cannons, Real-time earthquake data and simulated earthquake data.

6. The seismic full-waveform inversion method according to claim 5, characterized in that, The calculation of the gradient of the loss function with respect to the convolutional neural network model specifically includes: The derivatives of the loss function with respect to the first parameter, the second parameter, and the density are calculated using the adjoint state algorithm. Based on the density, the P-wave velocity, the S-wave velocity, and the derivatives of the loss function with respect to the first parameter and the second parameter, the derivatives of the loss function with respect to the P-wave velocity and the S-wave velocity are calculated using the chain rule. The gradient of the loss function with respect to the parameters in the convolutional neural network model is determined by calculating the derivatives of the loss function with respect to the P-wave velocity and the S-wave velocity using an automatic differentiation algorithm. The adjoint state algorithm is based on the cross-correlation between the forward wave field and the adjoint wave field, which includes the adjoint longitudinal normal stress wave field, the adjoint transverse normal stress wave field, and the adjoint shear stress wave field.

7. A seismic full-waveform inversion device, characterized in that, The device includes: The data acquisition module is used to input random vectors into a pre-constructed convolutional neural network model, output parameter variable data through the convolutional neural network model, and determine physical property parameter data based on the parameter variable data and pre-acquired initial parameter data; wherein, the parameter variable data includes P-wave velocity perturbation and S-wave velocity perturbation, and the initial parameter data includes initial P-wave velocity and initial S-wave velocity; The full waveform forward modeling module is used to perform forward modeling on the physical property parameter data to obtain simulated seismic data; The full waveform inversion module is used to calculate the residual value between the simulated seismic data and the pre-acquired actual seismic data. In response to the residual value being less than or equal to a preset threshold, the physical property parameter data is used as the target physical property parameter data of the seismic full waveform.

8. A seismic full-waveform inversion device according to claim 7, characterized in that, The physical property parameter data includes at least P-wave velocity and S-wave velocity, and the data acquisition module is also used for: Determine the physical property parameters using the following formula: ； In the formula, x represents a random vector. and These are the convolutional neural network models corresponding to the longitudinal wave velocity and the transverse wave velocity, respectively. and They are respectively and The parameters to be optimized in Represented as the initial P-wave velocity, Represented as the initial shear wave velocity, Expressed as longitudinal wave velocity, It is expressed as transverse wave velocity.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a seismic full-waveform inversion method as described in any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the seismic full waveform inversion method according to any one of claims 1 to 6.

Images