A full waveform inversion method and system for optimization acceleration

Abstract

The present invention provides an optimization-accelerated full waveform inversion method and system, belonging to the field of seismic data processing. The optimization-accelerated full waveform inversion method randomly extracts the original shot gather to obtain an extracted shot gather, and uses MPI to manage multiple groups of processes to obtain the optimal step size according to the extracted shot gather and different step sizes respectively, and finally obtains the velocity update amount by using the optimal step size. The present invention replaces the use of all calculation step sizes with randomly distributed shot gathers to improve the operation efficiency, without changing the original calculation steps and without affecting the original calculation effect. At the same time, by randomly selecting some shot gathers, the regularity is avoided and the statistical law is not affected. During the process of obtaining the step size, only some shot gathers are used to obtain the appropriate step size, and it is easy to implement through MPI management.

Description

Technical Field

[0001] The present invention belongs to the field of seismic data processing, and particularly relates to an optimization-accelerated full waveform inversion method and system. Background Art

[0002] The full waveform inversion technology is a highly non-linear inverse problem solving technology and a severely ill-posed problem. Therefore, a local linearization solution method is often adopted in the solving process. This means that there is a huge computational amount problem in full waveform inversion. In fact, the huge storage and computational amount have always been bottleneck problems restricting full waveform inversion. The practical application of FWI is necessarily related to the computational amount and computational efficiency. In a general iterative algorithm, at least three forward seismic wave calculations of all shot gathers are required for one iteration step of FWI. It can be seen from this that the computational amount of FWI is very huge.

[0003] Full waveform inversion is an effective tool for obtaining high-resolution subsurface parameters. This method is based on minimizing the difference between the observed values and the predicted values, and calculates the predicted values by solving the two-way wave propagation problem. Minimizing the difference between the predicted values and the observed values is actually a large-scale non-linear inversion problem. The common way to solve this method is gradient-based algorithms, such as non-linear conjugate gradient method, l-BFGS quasi-Newton, truncated Newton method, etc. These methods have shown different advantages in practical applications. However, in any case, multiple forward modeling operations are required to obtain an accurate step size, resulting in a large computational workload. Chinese Patent Publication No. CN103091711A discloses a full waveform inversion method and device based on the time-domain first-order velocity-stress elastic wave equation. The method includes: using the adjoint method based on perturbation theory to determine the adjoint equation of the time-domain first-order velocity-stress elastic wave equation and the gradient expression of the corresponding objective functional with respect to the model parameters; determining the forward-propagating wave field according to the time-domain first-order velocity-stress elastic wave equation, and determining the reverse-time extrapolated adjoint wave field according to the adjoint equation; determining the gradient of the objective functional with respect to the model parameters according to the forward-propagating wave field, the reverse-time extrapolated adjoint wave field, and the gradient expression; using a gradient-based iterative algorithm to perform multi-scale full waveform inversion. Chinese Patent Publication No. CN103592685A discloses a method and device for removing the direct wave in full waveform inversion. The method includes the following steps: inputting a velocity model; performing forward modeling of the velocity model based on the acoustic wave equation to obtain a simulated seismic record; calculating the time T0(x) when the direct wave reaches the surface receiver, where x is the distance from the geophone position to the source position; calculating the input time window T(x) according to the formula; deleting the seismic records received by the surface receiver whose reception time is less than the time T(x). Chinese Patent Publication No. CN103135132A discloses a hybrid-domain full waveform inversion method, which places the forward modeling part of full waveform inversion in the time domain, that is, performs forward modeling in the time domain and uses DFT to transform to the frequency domain for inversion, that is, uses the discrete Fourier transform to extract the wave field components corresponding to the inversion frequencies, and performs inversion from low frequency to high frequency in the frequency domain.

[0004] In recent years, with the proposal of some optimization algorithms and the rapid development of high-performance computing, the practical application process of full waveform inversion has been greatly promoted. However, no matter which algorithm is used, multiple forward modeling operations are required to find the optimal iteration step size. Summary of the Invention

[0005] The object of the present invention is to solve the problems existing in the above-mentioned prior art, and provide an optimized and accelerated full waveform inversion method and system. By simultaneously using different step sizes for forward modeling optimization of some randomly generated shot gathers, the computational efficiency can be greatly improved on the basis of the original algorithm.

[0006] The present invention is realized by the following technical solutions:

[0007] In the first aspect of the present invention, an optimization-accelerated full waveform inversion method is provided. The method randomly extracts the original shot gather to obtain an extracted shot gather, and uses MPI to manage multiple groups of processes to obtain the optimal step length according to the extracted shot gather and different step lengths respectively, and finally obtains the velocity update amount by using the optimal step length.

[0008] A further improvement of the present invention lies in that the method includes:

[0009] (1) Collect shot gather data and calculate the gradient;

[0010] (2) Randomly extract the original shot gather to obtain an extracted shot gather;

[0011] (3) Read the observation records of the extracted shot gather;

[0012] (4) Divide the processes into 3 groups through MPI, namely S1, S2 and S3;

[0013] (5) Assign different step lengths to each group of processes: the step lengths assigned to processes S1, S2 and S3 are α1, α2 and α3 respectively;

[0014] (6) The three processes perform calculations simultaneously to obtain three residuals of the extracted shot gather:

[0015] (7) Judge whether the step lengths satisfy the parabolic distribution. If so, go to step (8); if not, return to step (5);

[0016] (8) Calculate and obtain the optimal step length by using the three step lengths and the residuals;

[0017] (9) Obtain the velocity update amount by using the optimal step length.

[0018] A further improvement of the present invention lies in that the shot gather data in step (1) includes N single-shot records;

[0019] The operation of calculating the gradient in step (1) includes: calculating the gradient by using all the single-shot records in the shot gather data.

[0020] A further improvement of the present invention lies in that the operation of step (2) includes:

[0021] Use a random function to generate N / 3 positive integers between [0, N], and each positive integer represents a shot number;

[0022] Find the shots corresponding to these positive integers from the original shot gather, and these shots constitute the extracted shot gather.

[0023] A further improvement of the present invention lies in that the operation of step (6) includes:

[0024] Three processes S1, S2, and S3 simultaneously calculate using the gradients obtained in step (1) and their respective step sizes, and respectively obtain three residuals Res1, Res2, and Res3 of the extracted shot gathers.

[0025] A further improvement of the present invention lies in that the operation of judging whether the step size satisfies the parabolic distribution in step (7) includes:

[0026] Taking the step size α as the abscissa and the residual Res as the ordinate, fitting a parabola using the three points (α1, Res1), (α2, Res2), and (α3, Res3), and then judging whether these three points are on both sides of the perpendicular line of the parabola. If so, it is determined that the step size satisfies the parabolic distribution; if not, it is determined that the step size does not satisfy the parabolic distribution.

[0027] A further improvement of the present invention lies in that the operation of step (8) includes:

[0028] Obtaining the optimal step size using the following formula:

[0029]

[0030] A further improvement of the present invention lies in that the operation of step (9) includes:

[0031] Multiplying the optimal step size by the gradient to obtain the velocity update amount.

[0032] In the first aspect of the present invention, an optimization-accelerated full-waveform inversion system is provided, and the system includes:

[0033] A gradient calculation unit for collecting shot gather data and calculating the gradient;

[0034] An extraction unit connected to the gradient calculation unit for randomly extracting the original shot gather to obtain an extracted shot gather and reading the observation record of the extracted shot gather;

[0035] A process allocation unit connected to the extraction unit for dividing the processes into 3 groups, namely S1, S2, and S3, through MPI;

[0036] A step size allocation unit connected to the process allocation unit for allocating different step sizes to each group of processes;

[0037] A residual acquisition unit connected to the step size allocation unit for simultaneously calculating using three processes to obtain three residuals of the extracted shot gather:

[0038] A judgment unit connected to the step size allocation unit and the optimal step size acquisition unit respectively for judging whether the step size satisfies the parabolic distribution. If so, the optimal step size acquisition unit is started; if not, the step size allocation unit is started.

[0039] An optimal step size acquisition unit, which is respectively connected to the residual acquisition unit and the judgment unit, and is used to calculate and obtain the optimal step size by using three step sizes and residuals;

[0040] A speed update amount acquisition unit, which is connected to the optimal step size acquisition unit, and is used to obtain the speed update amount by using the optimal step size.

[0041] In a third aspect of the present invention, there is provided a computer-readable storage medium, where the computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer executes the steps in the optimization-accelerated full-waveform inversion method described above.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention replaces the use of all calculation step sizes with randomly distributed shot gathers to improve the operation efficiency, without changing the original calculation steps and without affecting the original calculation effect. At the same time, by randomly selecting some shot gathers, the regularity is avoided, and the statistical law is not affected. During the process of obtaining the step size, a suitable step size can be obtained by only using some shot gathers, and it is easy to implement through MPI management. Description of the Drawings

[0043] Figure 1 The step block diagram of the method of the present invention;

[0044] Figure 2 The structural schematic diagram of the system of the present invention. Detailed Embodiments

[0045] The present invention will be further described in detail below with reference to the drawings:

[0046] The present invention randomly re-samples the shot gathers to replace the forward optimization using all shot gathers, and solves the comparison problem of the actual records of the extracted shot gathers and the forward simulation data through MPI management. Three different iterative step sizes are assigned to three different processes by using MPI management, and a relative error is calculated respectively, and then a suitable step size is obtained. Under the condition of the same computing resources, the use of the present invention can greatly improve the computing efficiency without affecting the inversion accuracy.

[0047] An embodiment of an optimization-accelerated full-waveform inversion method of the present invention is as follows:

[0048]

Embodiment 1

[0049] As Figure 1 shown, the method of the present invention includes:

[0050] (1) Collect shot gather data, and use all single-shot records in the shot gather data to obtain the gradient g(v); the shot gather data includes N single-shot records;

[0051] The gradient is obtained by using the existing adjoint state method, that is, the forward propagation wave field is obtained by using all shot gather data (a total of N shot records), and the residuals of all shots are obtained. The residuals of all geophone points are used as the new backpropagation source to obtain the backpropagation wave field. The shot gather data is a collection of single-shot records in traditional seismic exploration, that is, a collection of records received by all geophone points in one excitation.

[0052] The formula for obtaining the gradient g(v) is as follows:

[0053]

[0054] In the formula: v is the model parameter; u is the forward propagation wave field of all shot points; xs is all single shots; t is the wave field propagation time; d obs is the observed record; B * represents the adjoint operator; R represents the limitation of the geophone point position; B * (R * (Ru - d obs )) represents the backpropagation wave field of the wave field residuals of all shot gathers in the model. Ru - dobs is the residual. Formula (1) is an existing formula and will not be elaborated here.

[0055] (2) Randomly extract the original shot gather to obtain the extracted shot gather:

[0056] Use a random function to generate N / 3 positive integers between [0, N]. If it cannot be divided evenly, round it up. Each positive integer represents a shot number, and the shots corresponding to these positive integers are found from the original shot gather. These shots constitute the extracted shot gather.

[0057] (3) Read the observed record d of the extracted shot gather obs : Obtain the observed record d of the extracted shot gather from the shot gather data obs .

[0058] (4) Divide the processes into 3 groups through MPI, namely S1, S2, and S3.

[0059] (5) Assign different step sizes to each group of processes:

[0060] Assign different step sizes to each group of processes. Specifically, the step sizes assigned to processes S1, S2, and S3 are α1, α2, and α3 respectively. The values of the step sizes α1, α2, and α3 can be set as needed. In this embodiment, the three step sizes are assigned the values of 0.01, 0.02, and 0.04 for the first time, and the three step sizes are assigned the values of 0.08, 0.16, and 0.32 for the second time, and so on, increasing the step size continuously. Other step sizes can also be adopted according to actual needs.

[0061] (6) Calculate the residuals of the extracted shot gather corresponding to different step sizes:

[0062] Three processes S1, S2, and S3 simultaneously calculate using the gradients obtained in step (1) and their respective step sizes, and respectively obtain the residuals Res1, Res2, and Res3 of the extracted gather. The method for calculating the residuals can adopt the existing method, which is briefly introduced as follows: Multiply the step size by the gradient to obtain the velocity update amount, update the initial velocity model with the velocity update amount to obtain a new velocity model, calculate the observed record using the new velocity model, and subtract the initial observed record d from the calculated observed record obs to obtain the residual Res.

[0063] In this step, the residuals of the extracted gather at three different step sizes are calculated simultaneously using three processes, replacing the original step of calculating three times, thus improving the operation efficiency.

[0064] (7) Determine whether the step sizes satisfy the parabolic distribution. If so, proceed to step (8). If not, return to step (5), that is, continue to assign values to the step sizes and repeat the above operations until three step sizes that satisfy the parabolic distribution are found.

[0065] The operation of determining whether the step sizes satisfy the parabolic distribution includes:

[0066] Taking the step size α as the abscissa and the residual Res as the ordinate, use the three points (α1, Res1), (α2, Res2), and (α3, Res3) to fit a parabola (the existing fitting method can be used and will not be elaborated here), and then determine whether these three points are on both sides of the perpendicular line of the parabola. If the three points are respectively on both sides of the perpendicular line (regardless of whether it is the two points on the left or the two points on the right), it is determined that the step sizes satisfy the parabolic distribution. If the three points are all on the same side of the perpendicular line, it is determined that the step sizes do not satisfy the parabolic distribution. The perpendicular line of the parabola is a straight line obtained by drawing a perpendicular line from the vertex or the lowest point of the parabola to the x-axis.

[0067] (8) Calculate the optimal step size using the three step sizes and the residuals;

[0068] The optimal step size is obtained using the following formula:

[0069]

[0070] (9) Calculate the velocity update amount using the optimal step size:

[0071] Calculate the velocity update amount for iteration using the optimal step size. The velocity update amount is the optimal step size multiplied by the gradient.

[0072] In full waveform inversion, after obtaining the velocity update, it is necessary to add the velocity update to the velocity model, and then repeat the process of obtaining the velocity update. The velocity model is updated using the newly obtained velocity update each time, that is, the iterative process. The number of iterations is determined according to the actual situation. The method of using the velocity update to complete full waveform inversion is a mature technology and will not be elaborated here. What the present invention aims to protect is the method for obtaining the velocity update.

[0073] The present invention also provides an optimized acceleration full waveform inversion system. Embodiments of the system are as follows:

[0074]

Embodiment 2

[0075] As Figure 2 shown, the system includes:

[0076] Gradient calculation unit 10, configured to collect shot gather data and calculate the gradient;

[0077] Extraction unit 20, connected to the gradient calculation unit 10, configured to randomly extract the original shot gather to obtain an extracted shot gather, and read the observation record of the extracted shot gather;

[0078] Process allocation unit 30, connected to the extraction unit 20, configured to divide the processes into three groups through MPI, namely S1, S2, and S3;

[0079] Step size allocation unit 40, connected to the process allocation unit 30, configured to allocate different step sizes to each group of processes;

[0080] Residual acquisition unit 50, connected to the step size allocation unit 40, configured to perform calculations simultaneously using three processes to obtain three residuals of the extracted shot gather:

[0081] Judgment unit 60, respectively connected to the step size allocation unit 40 and the optimal step size acquisition unit 70, configured to judge whether the step size satisfies the parabolic distribution. If so, the optimal step size acquisition unit 70 is started. If not, the step size allocation unit 40 is started;

[0082] Optimal step size acquisition unit 70, respectively connected to the residual acquisition unit 50 and the judgment unit 60, configured to calculate and obtain the optimal step size using three step sizes and residuals;

[0083] Velocity update acquisition unit 80, connected to the optimal step size acquisition unit 70, configured to obtain the velocity update using the optimal step size.

[0084] The present invention also provides a computer-readable storage medium. Embodiments of the computer-readable storage medium are as follows:

[0085]

Embodiment 3

[0086] The computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer is caused to execute the steps in the optimized acceleration full waveform inversion method:

[0087] (1) Collect shot gather data, and use all single-shot records in the shot gather data to obtain the gradient g(v); the shot gather data includes N single-shot records;

[0088] Use the existing adjoint state method to obtain the gradient, that is, use all shot gather data (a total of N shot records) to obtain the forward wave field, and obtain the residuals of all shots. Use the residuals of all geophone points as the new backpropagating source to obtain the backpropagating wave field. The shot gather data is a collection of single-shot records in traditional seismic exploration, that is, a collection of records received by all geophone points during one excitation.

[0089] The formula for obtaining the gradient g(v) is as follows:

[0090]

[0091] In the formula: v is the model parameter; u is the forward wave field of all shot points; xs is all single shots; t is the wave field propagation time; d obs is the observed record; B * represents the adjoint operator; R represents the limitation of the geophone point position; B * (R * (Ru - d obs )) represents the backpropagating wave field of the wave field residuals of all shot gathers in the model. Ru - dobs is the residual. Formula (1) is an existing formula and will not be elaborated here.

[0092] (2) Randomly extract the original shot gather to obtain the extracted shot gather:

[0093] Use a random function to generate N / 3 positive integers between [0, N]. If it cannot be divided evenly, round it up or down. Each positive integer represents a shot number, and find the shots corresponding to these positive integers from the original shot gather. These shots constitute the extracted shot gather.

[0094] (3) Read the observed record d of the extracted shot gather obs : Obtain the observed record d of the extracted shot gather from the shot gather data obs .

[0095] (4) Divide the processes into 3 groups through MPI, namely S1, S2, and S3.

[0096] (5) Assign different step sizes to each group of processes:

[0097] Assign different step sizes to each group of processes. Specifically, the step sizes assigned to processes S1, S2, and S3 are α1, α2, and α3 respectively. The values of step sizes α1, α2, and α3 can be set as needed. In this embodiment, for the first time, the three step sizes are respectively assigned the values of 0.01, 0.02, and 0.04. For the second time, the three step sizes are respectively assigned the values of 0.08, 0.16, and 0.32, and so on, continuously increasing the step sizes. Other step sizes can also be adopted according to actual needs.

[0098] (6) Calculate the residuals of the extracted shot gathers corresponding to different step sizes:

[0099] The three processes S1, S2, and S3 simultaneously use the gradients obtained in step (1) and their respective step sizes for calculation, and respectively obtain the residuals Res1, Res2, and Res3 of the extracted shot gathers. The method for calculating the residuals can adopt the existing method, which is briefly introduced as follows: Multiply the step size by the gradient to obtain the velocity update amount, update the initial velocity model with the velocity update amount to obtain a new velocity model, calculate the observed records using the new velocity model, and subtract the initial observed record d from the calculated observed records obs To obtain the residual Res.

[0100] In this step, the residuals of the extracted gather at three different step sizes are calculated simultaneously by three processes, replacing the original step of calculating three times, thus improving the operation efficiency.

[0101] (7) Determine whether the step sizes satisfy the parabolic distribution. If so, go to step (8). If not, return to step (5), that is, continue to assign values to the step sizes and repeat the above operations until three step sizes that satisfy the parabolic distribution are found.

[0102] The operation of determining whether the step sizes satisfy the parabolic distribution includes:

[0103] Taking the step size α as the abscissa and the residual Res as the ordinate, use the three points (α1, Res1), (α2, Res2), and (α3, Res3) to fit a parabola (the existing fitting method can be used and will not be elaborated here). Then determine whether these three points are on both sides of the perpendicular line of the parabola. If the three points are respectively on both sides of the perpendicular line (regardless of whether it is the two points on the left or the two points on the right), it is determined that the step sizes satisfy the parabolic distribution. If the three points are all on the same side of the perpendicular line, it is determined that the step sizes do not satisfy the parabolic distribution. The perpendicular line of the parabola is the straight line obtained by drawing a perpendicular line from the vertex or the lowest point of the parabola to the x-axis.

[0104] (8) Calculate the optimal step size using the three step sizes and the residuals;

[0105] The optimal step size is obtained using the following formula:

[0106]

[0107] (9) Calculate the velocity update using the optimal step size:

[0108] Obtain the velocity update for iteration using the optimal step size. The velocity update is the product of the optimal step size and the gradient.

[0109]

Embodiment 4

[0110] The full-waveform inversion has a huge computational cost. In addition to calculating the gradients of the forward and reverse wavefields for all shot gathers, multiple forward modeling operations are required for each iteration to obtain an appropriate step size. To calculate the gradients, all shot gathers must be used, but during the optimization process, a randomly selected subset of shot gathers can be used instead. Through MPI management, different shot gathers are used, and the observation records of these shot gathers are accurately recorded. During the forward modeling for optimization, only the errors of these shot gathers need to be compared to find an appropriate step size. At the same time, by randomly selecting a subset of shot gathers, regularity is avoided without affecting the statistical laws. During the process of obtaining the step size, an appropriate step size can be obtained using only a subset of shot gathers, and it is easily implemented through MPI management.

[0111] The computational cost is the bottleneck problem restricting full-waveform inversion. The present invention can effectively reduce the computational cost during the optimization process of full-waveform inversion, promote the practical application process of full-waveform inversion, and improve the operation efficiency by using randomly distributed shot gathers to replace the use of all computational steps without changing the original calculation steps or affecting the original calculation results. At the same time, by randomly selecting a subset of shot gathers, regularity is avoided without affecting the statistical laws. During the process of obtaining the step size, an appropriate step size can be obtained using only a subset of shot gathers, and it is easily implemented through MPI management. It has broad application prospects in the process of full-waveform inversion modeling and imaging.

[0112] Finally, it should be noted that the above technical solution is only one implementation manner of the present invention. For those skilled in the art, based on the disclosed application methods and principles of the present invention, various types of improvements or deformations can be easily made, not limited to the methods described in the above specific implementation manner of the present invention. Therefore, the above-described manner is only preferred and does not have a restrictive meaning.

Claims

1. An optimization-accelerated full-waveform inversion method, characterized in that: The method includes: (1) Collect shot gather data and calculate the gradient; (2) Randomly sample the original shot gather to obtain a sampled shot gather; (3) Read the observation records of the sampled shot gather; (4) Divide the processes into three groups, namely S1, S2, and S3 through MPI; (5) Assign different step sizes to each group of processes: the step sizes assigned to processes S1, S2, and S3 are α1, α2, and α3 respectively; (6) The three processes perform calculations simultaneously to obtain three residuals of the sampled shot gather: (7) Determine whether the step sizes satisfy the parabolic distribution. If so, proceed to step (8). If not, return to step (5); (8) Calculate the optimal step size using the three step sizes and residuals; (9) Obtain the velocity update amount using the optimal step size; The operations in step (6) include: The three processes S1, S2, and S3 perform calculations simultaneously using the gradient obtained in step (1) and their respective step sizes to obtain three residuals Res1, Res2, and Res3 of the sampled shot gather respectively; The operation of determining whether the step sizes satisfy the parabolic distribution in step (7) includes: Taking the step size α as the abscissa and the residual Res as the ordinate, fitting a parabola using the three points (α1, Res1), (α2, Res2), and (α3, Res3), and then determining whether these three points are on both sides of the perpendicular line of the parabola. If so, it is determined that the step sizes satisfy the parabolic distribution. If not, it is determined that the step sizes do not satisfy the parabolic distribution; The operations in step (8) include: Calculate the optimal step size using the following formula:

2. The full waveform inversion method for optimization acceleration according to claim 1, wherein: The shot gather data in step (1) includes N single-shot records; The operation of calculating the gradient in step (1) includes: calculating the gradient using all the single-shot records in the shot gather data.

3. The optimization-accelerated full waveform inversion method according to claim 2, characterized in that: The operations in step (2) include: Generate N / 3 positive integers between [0, N] using a random function, and each positive integer represents a shot number; Find the shots corresponding to these positive integers from the original shot gather, and these shots constitute the sampled shot gather.

4. The optimized acceleration full waveform inversion method according to claim 3, characterized in that: The operations in step (9) include: Multiply the optimal step size by the gradient to obtain the velocity update amount.

5. An all-waveform inversion system for optimization acceleration, characterized in that: The system includes: A gradient calculation unit for collecting shot gather data and calculating the gradient; A sampling unit connected to the gradient calculation unit for randomly sampling the original shot gather to obtain a sampled shot gather and reading the observation records of the sampled shot gather; A process allocation unit connected to the sampling unit for dividing the processes into three groups, namely S1, S2, and S3 through MPI; A step size allocation unit connected to the process allocation unit for assigning different step sizes to each group of processes; A residual acquisition unit connected to the step size allocation unit for performing calculations simultaneously using the three processes to obtain three residuals of the sampled shot gather: A judgment unit connected to the step size allocation unit and the optimal step size acquisition unit respectively for determining whether the step sizes satisfy the parabolic distribution. If so, activate the optimal step size acquisition unit. If not, activate the step size allocation unit; An optimal step size acquisition unit connected to the residual acquisition unit and the judgment unit respectively for calculating the optimal step size using the three step sizes and residuals; A speed update amount acquisition unit, connected to the optimal step size acquisition unit, is configured to obtain a speed update amount by using the optimal step size.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer is caused to execute the steps in the optimization acceleration full waveform inversion method according to any one of claims 1 to 4.

Images