Full waveform inversion method and device based on deep learning optimization

Abstract

The invention discloses a full waveform inversion method and device based on deep learning optimization. The method comprises the following steps: inputting seismic data; calculating a gradient in a full-wave inversion process by using a deep learning model; and outputting a full-waveform inversion result. According to the invention, the deep learning technology is applied to full-waveform inversion, the gradient in the full-waveform inversion process is approximately calculated through the deep learning model, and a full-waveform inversion alternative scheme with high calculation efficiency, strong robustness and good adaptability is provided. Therefore, the high calculation cost of solving forward modeling and adjoint equations each time in traditional full waveform inversion is avoided, and the efficiency and accuracy of inversion are improved.

Description

Technical Field

[0001] The present invention relates to the field of geophysics and artificial intelligence technology, and in particular to a full waveform inversion method and device based on deep learning optimization. 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 residual of seismic data. The core of FWI technology is to solve the wave equation to simulate the propagation of seismic waves in the underground medium and compare it with the actual observation data. Although FWI can provide high-resolution underground structure images, its calculation cost is high and time-consuming, especially when processing large-scale three-dimensional seismic data. Existing FWI methods mainly rely on iterative optimization algorithms, such as gradient descent, to update the underground velocity model. These methods require calculating 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: High computational cost: The solution of the adjoint state equation requires a lot of computing resources, especially when dealing with large-scale data sets, which limits the application scope of FWI technology.

[0003] Slow convergence speed: The traditional FWI method 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

[0004] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art and provide a full waveform inversion method and device based on deep learning optimization to reduce computing costs and improve inversion efficiency and accuracy.

[0005] To achieve the above object, the technical solution of the present invention is: In a first aspect, the present invention provides a full waveform inversion method based on deep learning optimization, comprising: Input earthquake data; Use deep learning models to calculate gradients during full-wave inversion; Output full waveform inversion results.

[0006] Optionally, the formula for full waveform inversion is:

[0007] 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 It's time. f(u,m,t) is the misfit function, which is the square of the residual.

[0008] Optionally, the method of using the deep learning model to calculate the gradient in the full-wave inversion process is:

[0009] in, is the loss function L Relative to medium parameters m The gradient of It is the parameter learned by the deep learning model and is used for approximate gradient calculation; It is a simulated wave field u Medium parameters m sensitivity; d is the observation data.

[0010] Optionally, the full waveform inversion method based on deep learning optimization also includes: The full waveform inversion results are compared with the results of the subsurface velocity model.

[0011] In a second aspect, the present invention provides a full waveform inversion device based on deep learning optimization, comprising: An input module, used to input seismic data; A deep learning module for calculating gradients during full-wave inversion using a deep learning model; Output module, used to output full waveform inversion results.

[0012] Compared with the prior art, the present invention has the following beneficial effects: The present invention applies deep learning technology to full waveform inversion, and uses a deep learning model to approximate the calculation of the gradient in the full-wave inversion process, providing a full-wave inversion alternative 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 the inversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flowchart of a full waveform inversion method based on deep learning optimization provided in an embodiment of the present application; Figure 2 A flowchart of a full waveform inversion method based on deep learning optimization provided in a preferred embodiment of the present application; Figure 3 A schematic diagram of the composition of an electronic device provided in an embodiment of the present application; Figure 4 It is the operation interface of the Matlab program; Figure 5A schematic diagram of the composition of a full waveform inversion device based on deep learning optimization provided in an embodiment of the present application; Figure 6 A schematic diagram of the composition of a full waveform inversion device based on deep learning optimization provided in a preferred embodiment of the present application; The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0016] Embodiment 1: See also Figure 1 As shown, the full waveform inversion method based on deep learning optimization provided in this embodiment mainly includes the following steps: 110, input earthquake data; 120, using a deep learning model to compute gradients during full-wave inversion; In this step, the deep learning model can quickly obtain the 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 computing resources and improving computing 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, the convergence speed of the full-wave inversion can be accelerated. The deep learning model can directly provide an approximate value of the gradient, simplifying the iterative process of the full-wave inversion and reducing storage requirements.

[0017] 130, output the full waveform inversion result.

[0018] It can be seen that this method applies deep learning technology to full waveform inversion, and uses a deep learning model to approximate the gradient in the full waveform inversion process, thereby avoiding the high computational cost of solving the adjoint state equation in traditional full waveform inversion.

[0019] Specifically, full waveform inversion is a seismic imaging method based on wave equation constraints, which aims to generate a high-resolution subsurface model by iteratively minimizing the mismatch between simulated and observed seismograms, as follows:

[0020] in, 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 It's time. f(u,m,t) is the misfit function, usually the square of the residual.

[0021] In equation solving, the adjoint state method is a commonly used method for calculating gradients. The adjoint state method involves the definition of the adjoint wave field, which satisfies the following differential equation:

[0022] in, is the companion wave field, u is the forward wave field, d is the observed data, is the Dirac function, is the epicenter location.

[0023] The gradient operator formula for full waveform inversion in the time domain based on the adjoint state method is:

[0024] in, is the gradient of the objective function with respect to the medium parameters, is the sensitivity of the wave field to medium parameters.

[0025] However, the solution of the adjoint state equation requires a lot of computing resources, especially when processing large-scale data sets, which limits the scope of application of full waveform inversion technology. To this end, the embodiment of the present application uses a deep learning model to calculate the gradient in the full wave inversion process and directly predict the underground velocity model from the seismic data. Specifically, the gradient calculated by using the deep learning model in the full wave inversion process can be expressed as:

[0026] in, is the loss function L Relative to medium parameters m gradient. It is the parameter learned by the deep learning model and is used for approximate gradient calculation. It's TRON u Medium parameters m u is the simulated wave field and d is the observed data.

[0027] Specifically, the parameters It is obtained in the following way: the deep learning model takes the seismic data as input and the underground velocity model corresponding to the seismic data as output, uses a loss function to measure the gap between the predicted underground velocity model and the actual underground velocity model, uses the gradient descent algorithm for back propagation, calculates the gradient of the loss function to the deep learning model parameters, and updates the parameters according to this gradient. , through multiple iterations of training and continuous adjustment, the final parameters are obtained , so that the underground velocity model can be accurately predicted from seismic data. To approximate the gradient, a computationally efficient, robust, and adaptable alternative to FWI is provided, thereby avoiding the high computational cost of solving the forward and adjoint equations each time in traditional FWI and improving the efficiency and accuracy of the inversion.

[0028] In a preferred embodiment, the full waveform inversion method based on deep learning optimization further includes the following steps: 140, comparing the full waveform inversion result with the result of the real underground velocity model.

[0029] Through this step, the accuracy of the full waveform inversion method based on deep learning optimization in this embodiment can be determined.

[0030] like Figure 3 As shown, the 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, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other 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 based on deep learning optimization provided in the above embodiment when executing the program stored in the memory 113. Figure 4 The figure shows the operation interface of the MATLAB program of the electronic device, including the input and output of seismic data and logging data, deep learning model training, FWI execution and result display; the blue circles in area A and area B are the real underground velocity model data, the red stars are the FWI inversion data, area A is the comparison chart of the traditional FWI inversion results and the real underground velocity model data, and area B is the comparison chart of the deep learning FWI inversion results using this method and the velocity model data. It can be seen that there is a certain deviation between the traditional FWI inversion results and the real underground velocity model data, while the deep learning FWI inversion results of this method are in good agreement with the real underground velocity model data, and the inversion time is shortened to 1 / 5 of the original time, which greatly improves the inversion efficiency. Overall, the accuracy and efficiency of the inversion results are significantly improved.

[0031] Accordingly, this embodiment provides a full waveform inversion device based on deep learning optimization, such as Figure 5 As shown, including: An input module 510, for inputting seismic data; A deep learning module 520 for using a deep learning model to calculate gradients during full-wave inversion to directly predict a subsurface velocity model from seismic data; The output module 530 is used to output the inversion result of the full waveform inversion.

[0032] The goal of the full waveform inversion is to generate a subsurface velocity model by iteratively minimizing the mismatch between simulated and observed seismic maps. The formula for the full waveform inversion is:

[0033] 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 It's time. f(u,m,t) is the misfit function, which is the square of the residual.

[0034] The method of using the deep learning model to calculate the gradient in the full-wave inversion process is:

[0035] in, is the loss function L Relative to medium parameters m The gradient of It is the parameter learned by the deep learning model and is used for approximate gradient calculation; It is a simulated wave field u Medium parameters m sensitivity; d is the observation data.

[0036] In a preferred embodiment, Figure 6 As shown, the full waveform inversion device based on deep learning optimization also includes: The comparison module 540 is used to compare the full waveform inversion result with the result of the real underground velocity model.

[0037] It should be noted that the full waveform inversion device based on deep learning optimization provided in the embodiment of the present application can execute the full waveform inversion method based on deep learning optimization provided in any embodiment of the present application, and has the corresponding functions and beneficial effects of the execution method.

[0038] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made based on the essence of the content of the present invention should be included in the protection scope of the present invention.

Claims

1. A full waveform inversion method based on deep learning optimization, characterized in that: include: Input earthquake data; Use deep learning models to calculate gradients during full-wave inversion; Output full waveform inversion results; The method of using the deep learning model to calculate the gradient in the full-wave inversion process is: ； in, is the loss function L Relative to medium parameters m The gradient of It is the parameter learned by the deep learning model and is used for approximate gradient calculation; It is a simulated wave field u Medium parameters m sensitivity; d is the observation data.

2. The full waveform inversion method based on deep learning optimization according to claim 1, characterized in that: The formula for the full waveform inversion is: ； 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 It's time. f (u,m,t) is the misfit function, which is the square of the residual.

3. The full waveform inversion method based on deep learning optimization according to claim 2, characterized in that: parameter Obtained by: The deep learning model takes seismic data as input and the underground velocity model corresponding to the seismic data as output. It uses the loss function to measure the gap between the predicted underground velocity model and the actual underground velocity model. It uses the gradient descent algorithm for back propagation, calculates the gradient of the loss function to the deep learning model parameters, and updates the parameters according to the gradient. , through multiple iterations of training and continuous adjustment, the final parameters are obtained .

4. The full waveform inversion method based on deep learning optimization according to claim 1, characterized in that: Also includes: The full waveform inversion results are compared with the results of the true subsurface velocity model.

5. A full waveform inversion device based on deep learning optimization, characterized in that: include: An input module, used to input seismic data; A deep learning module for calculating gradients during full-wave inversion using a deep learning model; Output module, used for outputting full waveform inversion results; The method of using the deep learning model to calculate the gradient in the full-wave inversion process is: ； in, is the loss function L Relative to medium parameters m The gradient of It is the parameter learned by the deep learning model and is used for approximate gradient calculation; It is a simulated wave field u Medium parameters m sensitivity; d is the observation data.

6. The full waveform inversion device based on deep learning optimization according to claim 5, characterized in that: The formula for the full waveform inversion is: ； 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 It's time. f (u,m,t) is the misfit function, which is the square of the residual.

7. The full waveform inversion device based on deep learning optimization according to claim 5, characterized in that: parameter Obtained by: The deep learning model takes seismic data as input and the underground velocity model corresponding to the seismic data as output. It uses a loss function to measure the gap between the predicted underground velocity model and the actual underground velocity model. It uses the gradient descent algorithm for back propagation, calculates the gradient of the loss function to the deep learning model parameters, and updates the parameters according to the gradient. , through multiple iterations of training and continuous adjustment, the final parameters are obtained .

8. The full waveform inversion device based on deep learning optimization according to claim 5, characterized in that: Also includes: The comparison module is used to compare the full waveform inversion result with the result of the underground velocity model.

Images