A full waveform inversion method and device optimized based on deep learning

Abstract

The present invention discloses a full waveform inversion method and device optimized based on deep learning. The method includes inputting seismic data; using a deep learning model to calculate the gradient in the full wave inversion process; and outputting the full waveform inversion result. The present invention applies deep learning technology to full waveform inversion, approximates the calculation of the gradient in the full waveform inversion process through a deep learning model, provides an alternative full waveform inversion solution with high computational efficiency, strong robustness, and good adaptability, thereby avoiding the high computational cost of solving the forward and adjoint equations each time in traditional full waveform inversion, and improving the efficiency and accuracy of inversion.

Description

Technical Field

[0001] The present invention relates to the fields of geophysics and artificial intelligence technology, and particularly relates to a full waveform inversion method and device optimized based on deep learning. Background Art

[0002] In the field of geophysics, full waveform inversion (FWI) is an important seismic imaging technology that constructs an underground velocity model by minimizing the residuals of seismic data. The core of the FWI technology lies in solving the wave equation to simulate the propagation of seismic waves in the underground medium and comparing it with the actual observed data. Although FWI can provide high-resolution images of underground structures, its computational cost is high and time-consuming, especially when dealing with large-scale three-dimensional seismic data. Existing FWI methods mainly rely on iterative optimization algorithms, such as the gradient descent method, to update the underground velocity model. These methods need to calculate the sensitivity of the simulated wave field to the velocity model, that is, the gradient, which is usually achieved by solving the adjoint state equation. However, this method has the following disadvantages:

[0003] High computational cost: Solving the adjoint state equation requires a large amount of computational resources, especially when dealing with large-scale data sets, which limits the application scope of the FWI technology.

[0004] Slow convergence speed: Traditional FWI methods may require hundreds or even thousands of iterations to converge to a reasonable velocity model, which makes the entire inversion process time-consuming and inefficient. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above-mentioned deficiencies of the prior art, and provide a full waveform inversion method and device optimized based on deep learning to reduce the computational cost and improve the inversion efficiency and accuracy.

[0006] To achieve the above purpose, the technical solution of the present invention is:

[0007] In the first aspect, a full waveform inversion method optimized based on deep learning according to the present invention includes:

[0008] Input seismic data;

[0009] Use a deep learning model to calculate the gradient in the full wave inversion process;

[0010] Output the full waveform inversion result.

[0011] Optionally, the formula for the full waveform inversion is:

[0012]

[0013] J(m)is the objective function that measures the simulated wavefield u and the difference between the observed wavefield; m is the medium parameter, t is the time, f(u,m,t) is the misfit function, which is the square of the residual.

[0014] Optionally, the method of using a deep learning model to calculate the gradient in the full-wave inversion process is as follows:

[0015]

[0016] wherein, is the loss function L with respect to the medium parameter m gradient; are the parameters learned by the deep learning model for approximating gradient calculation; is the simulated wavefield u with respect to the medium parameter m sensitivity; d is the observed data.

[0017] Optionally, the full-waveform inversion method based on deep learning optimization further includes:

[0018] Comparing the full-waveform inversion result with the result of the subsurface velocity model.

[0019] In a second aspect, the present invention provides a full-waveform inversion device based on deep learning optimization, including:

[0020] An input module for inputting seismic data;

[0021] A deep learning module for using a deep learning model to calculate the gradient in the full-wave inversion process;

[0022] An output module for outputting the full-waveform inversion result.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] The present invention applies deep learning technology to full-waveform inversion and approximates the calculation of the gradient in the full-wave inversion process through a deep learning model, providing an alternative full-wave inversion solution with high computational efficiency, strong robustness, and good adaptability, thereby avoiding the high computational cost of solving the forward and adjoint equations each time in traditional full-wave inversion and improving the efficiency and accuracy of inversion. Description of the Drawings

[0025] Figure 1 is a flowchart of the full-waveform inversion method based on deep learning optimization provided by an embodiment of the present application;

[0026] Figure 2 Flowchart of the full waveform inversion method optimized based on deep learning provided by the preferred embodiment of the present application;

[0027] Figure 3 Schematic diagram of the composition of the electronic device provided by the embodiment of the present application;

[0028] Figure 4 Matlab program operation interface;

[0029] Figure 5 Schematic diagram of the composition of the full waveform inversion device optimized based on deep learning provided by the embodiment of the present application;

[0030] Figure 6 Schematic diagram of the composition of the full waveform inversion device optimized based on deep learning provided by the preferred embodiment of the present application;

[0031] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings. Specific embodiments

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts fall within the scope of protection of the present application.

[0034] Embodiment 1:

[0035] Refer to Figure 1 As shown, the full waveform inversion method optimized based on deep learning provided by this embodiment mainly includes the following steps:

[0036] 110. Input seismic data;

[0037] 120. Use a deep learning model to calculate the gradient during the full wave inversion process;

[0038] In this step, the deep learning model can quickly obtain an approximate gradient, avoiding the high computational cost of solving the forward and adjoint equations in each iteration of the traditional full-wave inversion process, significantly reducing the demand for computational resources, and improving computational efficiency. At the same time, the generalization ability of the deep learning model helps to improve the stability and reliability of the inversion. Using the prediction ability of the deep learning model can accelerate the convergence speed of the full-wave inversion. The deep learning model can directly provide an approximate value of the gradient, simplifying the iterative process of the full-wave inversion and reducing the storage requirements.

[0039] 130, output the full-waveform inversion result.

[0040] It can be seen that this method applies deep learning technology to full-waveform inversion. By using the deep learning model to approximately calculate the gradient in the full-waveform inversion process, it avoids the high computational cost of solving the adjoint state equation in traditional full-waveform inversion.

[0041] Specifically, full-waveform inversion is a seismic imaging method based on wave equation constraints. Its goal is to generate a high-resolution subsurface model by iteratively minimizing the mismatch between the simulated and observed seismograms. The formula is as follows:

[0042]

[0043] where J(m) is the objective function, which measures the difference between the simulated wavefield u and the observed wavefield; m is the medium parameter, t is the time, f(u,m,t) is the mismatch function, usually the square of the residual.

[0044] In equation solving, the adjoint state method is a commonly used method for calculating the gradient. The adjoint state method involves the definition of the adjoint wavefield, which satisfies the following differential equation:

[0045]

[0046] where is the adjoint wavefield, u is the forward wavefield, d is the observed data, is the Dirac function, is the source location.

[0047] The formula for the time-domain full-waveform inversion gradient operator based on the adjoint state method is:

[0048]

[0049] where is the gradient of the objective function with respect to the medium parameter, is the sensitivity of the wavefield to the medium parameter.

[0050] However, solving the adjoint state equation requires a large amount of computing resources, especially when dealing with large-scale data sets, which limits the application scope of the full-waveform inversion technology. For this reason, the embodiments of the present application use a deep learning model to calculate the gradient in the full-wave inversion process and directly predict the subsurface velocity model from seismic data. Specifically, using a deep learning model to calculate the gradient in the full-wave inversion process can be expressed as:

[0051]

[0052] Wherein, is the loss function L with respect to the medium parameter m gradient. is the parameter learned by the deep learning model and is used to approximate the gradient calculation. is the wave field u with respect to the medium parameter m sensitivity. u is the simulated wave field and d is the observed data.

[0053] Specifically, the parameter is obtained in the following way: the deep learning model takes seismic data as input and the subsurface velocity model corresponding to the seismic data as output, uses a loss function to measure the gap between the predicted subsurface velocity model and the true subsurface velocity model, adopts the gradient descent algorithm for backpropagation, calculates the gradient of the loss function with respect to the parameters of the deep learning model, and updates the parameters according to this gradient . Through multiple iterative trainings and continuous adjustments, the final parameter is obtained, so that the subsurface velocity model can be accurately predicted from seismic data. In this way, the deep learning model approximates the gradient through the learned parameter , providing a computationally efficient, robust, and adaptable FWI alternative, thus avoiding the high computational cost of solving the forward and adjoint equations each time in traditional FWI and improving the efficiency and accuracy of inversion.

[0054] In a preferred embodiment, the full-waveform inversion method optimized based on deep learning further includes the following steps:

[0055] 140. Compare the full-waveform inversion result with the result of the true subsurface velocity model.

[0056] Through this step, the accuracy of the full-waveform inversion method optimized based on deep learning in this embodiment can be judged.

[0057] Such as Figure 3As shown in the figure, an embodiment of the present application provides an electronic device, including a processor 111, a communication interface 112, a memory 113, and a communication bus 114. Among them, the processor 111, the communication interface 112, and the memory 113 complete mutual communication through the communication bus 114; the memory 113 is used to store computer programs; the processor 111 is used to implement the steps of the full waveform inversion method optimized based on deep learning provided in the above embodiment when executing the programs stored on the memory 113. As Figure 4 shown, it is the matlab program operation interface of the electronic device, including the input and output of seismic data and logging data, deep learning model training, FWI execution, and result display; among them, the blue circles in area A and area B are the true underground velocity model data, and the red stars are the FWI inversion data. Area A is a comparison chart of the traditional FWI inversion result and the true underground velocity model data, and area B is a comparison chart of the deep learning FWI inversion result using this method and the velocity model data. It can be seen that there is a certain deviation between the traditional FWI inversion result and the true underground velocity model data, while the deep learning FWI inversion result using this method fits well with the true underground velocity model data. Moreover, the inversion time is shortened to 1 / 5 of the original, greatly improving the inversion efficiency. Overall, the accuracy and efficiency of the inversion result are significantly improved.

[0058] Correspondingly, this embodiment provides a full waveform inversion device optimized based on deep learning, as Figure 5 shown, including:

[0059] An input module 510, configured to input seismic data;

[0060] A deep learning module 520, configured to use a deep learning model to calculate the gradient in the full wave inversion process to directly predict the underground velocity model from seismic data;

[0061] An output module 530, configured to output the inversion result of the full waveform inversion.

[0062] The full waveform inversion target is to generate an underground velocity model by iteratively minimizing the mismatch between the simulated and observed seismograms. The formula for the full waveform inversion is:

[0063]

[0064] J(m) is the objective function, which measures the simulated wave field u and the difference between the observed wave fields; m is the medium parameter, t is the time, f(u,m,t) is the mismatch function, which is the square of the residual.

[0065] The method of using a deep learning model to calculate the gradient in the full-wave inversion process is as follows:

[0066]

[0067] where, is the loss function L with respect to the medium parameter m gradient; are the parameters learned by the deep learning model for approximating gradient calculation; is the simulated wavefield u with respect to the medium parameter m sensitivity; d is the observed data.

[0068] In a preferred embodiment, as Figure 6 shown, the full-waveform inversion device optimized based on deep learning further includes:

[0069] A comparison module 540 for comparing the full-waveform inversion result with the result of the true subsurface velocity model.

[0070] It should be noted that the full-waveform inversion device optimized based on deep learning provided in the embodiments of the present application can execute the full-waveform inversion method optimized based on deep learning provided in any embodiment of the present application, and has the corresponding functions and beneficial effects of executing the method.

[0071] The above embodiments are only used to illustrate the technical concept and characteristics of the present invention, and the purpose is to enable those of ordinary skill in the art to understand the content of the present invention and implement it accordingly, and cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the essence of the content of the present invention should be covered within the protection scope of the present invention.

Claims

1. A full-waveform inversion method optimized based on deep learning, characterized in that, Comprising: Input seismic data; Using a deep learning model to calculate the gradient in the full-wave inversion process; Outputting the full-waveform inversion result; The way of using the deep learning model to calculate the gradient in the full-wave inversion process is: Among them, is the loss function L with respect to the medium parameter m gradient; are the parameters learned by the deep learning model for approximating gradient calculation; is the simulated wavefield u with respect to the medium parameter m sensitivity; d is the observed data; Parameter obtained in the following manner: The deep learning model takes seismic data as input and the corresponding subsurface velocity model of the seismic data as output. The loss function is used to measure the gap between the predicted subsurface velocity model and the true subsurface velocity model. The gradient descent algorithm is adopted for backpropagation to calculate the gradient of the loss function with respect to the parameters of the deep learning model, and the parameters are updated according to the gradient , and through multiple iterative trainings and continuous adjustments, the final parameters are obtained .

2. The full waveform inversion method optimized based on deep learning according to claim 1, wherein The formula for the full-waveform inversion is: J(m) is the objective function that measures the simulated wavefield u and the difference between the observed wavefield; m is the medium parameter, t is the time, f (u,m,t) is the misfit function, which is the square of the residual.

3. The full waveform inversion method optimized based on deep learning according to claim 1, characterized in that, Further comprising: Comparing the full-waveform inversion result with the result of the true subsurface velocity model.

4. A full waveform inversion device optimized based on deep learning, characterized in that Comprising: An input module for inputting seismic data; A deep learning module for using a deep learning model to calculate the gradient in the full-wave inversion process; An output module for outputting the full-waveform inversion result; The way of using the deep learning model to calculate the gradient in the full-wave inversion process is: Among them, is the loss function L with respect to the medium parameter m gradient; is the parameter learned by the deep learning model for approximating the gradient calculation; is the simulated wavefield u with respect to the medium parameter m sensitivity; d is the observed data; Parameter obtained in the following manner: The deep learning model takes seismic data as input and the corresponding subsurface velocity model of the seismic data as output. A loss function is used to measure the gap between the predicted subsurface velocity model and the true subsurface velocity model. The gradient descent algorithm is adopted for backpropagation to calculate the gradient of the loss function with respect to the parameters of the deep learning model, and the parameters are updated according to the gradient , and through multiple iterative trainings and continuous adjustments, the final parameters are obtained .

5. The full waveform inversion device optimized based on deep learning according to claim 4, characterized in that, The formula for the full-waveform inversion is: J(m) is the objective function that measures the simulated wave field u and the difference between the observed wave field; m is the medium parameter, t is the time, f (u,m,t) is the misfit function, which is the square of the residual.

6. The full waveform inversion device optimized based on deep learning according to claim 4, characterized in that, Further comprising: A comparison module for comparing the full-waveform inversion result with the result of the subsurface velocity model.

Images