System and method for generating high-quality seismic data from low-quality seismic data

1 . A method for generating high-quality seismic data, the method comprising:

receiving a first dataset;

selecting a first processing technique based upon the first dataset;

producing first processed data from the first dataset using the first processing technique;

selecting a second processing technique based upon the first dataset, wherein the second processing technique is different than the first processing technique;

producing second processed data from the first dataset using the second processing technique;

training a machine learning model to transform the first processed data into the second processed data;

receiving a second dataset;

transforming the second dataset into a third dataset using the trained machine learning model; and

performing a wellsite action based upon the third dataset,

wherein:

the wellsite action comprises at least one of generating or transmitting a signal that instructs or causes a physical action to occur at a wellsite, and

the physical action comprises:

selecting where to drill a wellbore;

drilling the wellbore;

varying at least one of a weight or a torque on a drill bit that is drilling the wellbore;

varying a drilling trajectory of the wellbore;

varying at least one of a concentration or a flow rate of a fluid pumped into the wellbore; or

a combination thereof.

2 . The method of claim 1 , wherein the first dataset comprises seismic data that represents a subsurface.

3 . The method of claim 1 , wherein the first processed data has a higher quality than the first dataset.

4 . The method of claim 1 , wherein the second processed data has a higher quality than the first dataset and the first processed data.

5 . The method of claim 1 , wherein the third dataset has a higher quality than the second dataset.

6 . The method of claim 1 , further comprising displaying the third dataset.

7 . The method of claim 1 , further comprising creating or updating a model of a subsurface based upon the third dataset.

8 . A computing system for generating high-quality seismic data from low-quality seismic data, the computing system comprising:

one or more processors; and

a memory system including one or more non-transitory computer-readable media storing instructions that, when executed by at least one of the one or more processors, cause the computing system to perform operations, the operations including:

receiving a first dataset, wherein the first dataset comprises seismic data that represents a subsurface;

selecting a first processing technique based upon the first dataset;

producing first processed data from the first dataset using the first processing technique, wherein the first processed data has a higher quality than the first dataset;

selecting a second processing technique based upon the first dataset, wherein the second processing technique is different than the first processing technique;

producing second processed data from the first dataset using the second processing technique, wherein the second processed data has a higher quality than the first dataset and the first processed data;

training a machine learning model to transform the first processed data into the second processed data;

receiving a second dataset;

transforming the second dataset into a third dataset using the trained machine learning model, wherein the third dataset has a higher quality than the first dataset and the second dataset; and

performing a wellsite action based upon the third dataset,

wherein:

the wellsite action comprises at least one of generating or transmitting a signal that instructs or causes a physical action to occur at a wellsite, and

the physical action comprises:

selecting where to drill a wellbore;

drilling the wellbore;

varying at least one of a weight or a torque on a drill bit that is drilling the wellbore;

varying a drilling trajectory of the wellbore;

varying a concentration and/or flow rate of a fluid pumped into the wellbore; or

a combination thereof.

9 . The computing system of claim 8 , wherein the seismic data comprises particle data, and wherein the particle data comprises particle displacement data, particle velocity data, particle acceleration data, or a combination thereof.

10 . The computing system of claim 8 , wherein the first and second processing techniques comprise denoising, deblending, interpolating, deconvoluting, demultiple, deghosting, imaging, inversion, residual moveout correction (RMO), noise suppression, regularization, or a combination thereof.

11 . The computing system of claim 8 , wherein the machine learning model comprises a generative adversarial network, a variational autoencoder, a diffusion model, a transformer, a neural radiance field, a convolution-based model, a non-convolution-based model, or a combination thereof.

12 . The computing system of claim 8 , wherein training comprises a generator loss function, a discriminator loss function, an adversarial loss function, or a combination thereof.

13 . A non-transitory computer-readable medium storing instructions for autonomously performing a subsurface operation that, when executed by one or more processors of a computing system, cause the computing system to perform operations, the operations comprising:

receiving a first dataset, wherein the first dataset comprises seismic data that represents a subsurface, wherein the seismic data is from a land or marine environment, wherein the seismic data comprises particle data, and wherein the particle data comprises particle displacement data, particle velocity data, particle acceleration data, or a combination thereof;

selecting a first processing technique based upon the first dataset, wherein the first processing technique comprises denoising, deblending, interpolating, deconvoluting, demultiple, deghosting, imaging, inversion, residual moveout correction (RMO), noise suppression, regularization, or a combination thereof;

producing first processed data from the first dataset using the first processing technique, wherein the first processed data has a higher quality than the first dataset;

selecting a second processing technique based upon the first dataset, wherein the second processing technique comprises denoising, deblending, interpolating, deconvoluting, demultiple, deghosting, imaging, inversion, RMO, noise suppression, regularization, or a combination thereof, and wherein the second processing technique is different than the first processing technique;

producing second processed data from the first dataset using the second processing technique, wherein the second processed data has a higher quality than the first dataset and the first processed data;

training a machine learning model to transform the first processed data into the second processed data, wherein the machine learning model comprises a generative adversarial network, a variational autoencoder, a diffusion model, a transformer, a neural radiance field, a convolution-based model, a non-convolution-based model, or a combination thereof, and wherein training comprises or generates a generator loss function, a discriminator loss function, an adversarial loss function, a modified objective function, or a combination thereof;

receiving a second dataset;

transforming the second dataset into a third dataset using the trained machine learning model, wherein the third dataset has a higher quality than the first dataset and the second dataset;

displaying the third dataset; and

performing a wellsite action based upon the third dataset,

wherein:

the wellsite action comprises at least one of generating or transmitting a signal that instructs or causes a physical action to occur at a wellsite, and

the physical action comprises:

selecting where to drill a wellbore;

drilling the wellbore;

varying at least one of a weight or a torque on a drill bit that is drilling the wellbore;

varying a drilling trajectory of the wellbore;

varying at least one of a concentration or a flow rate of a fluid pumped into the wellbore; or

a combination thereof.

14 . The non-transitory computer-readable medium of claim 13 , wherein the first processing technique comprises a first interpolation technique, wherein the second processing technique comprises a second interpolation technique, and wherein the first interpolation technique is lower resolution than the second interpolation technique.

15 . The non-transitory computer-readable medium of claim 14 , wherein the first interpolation technique comprises Fourier-based interpolation, and wherein the second interpolation technique comprises Radon-based interpolation.

16 . The non-transitory computer-readable medium of claim 13 , wherein the operations further comprise creating or updating a model of the subsurface based upon the third dataset.

17 . The non-transitory computer-readable medium of claim 13 , wherein the operations further comprise performing a wellsite action based upon the third dataset, and wherein the wellsite action comprises generating and/or transmitting a signal that instructs or causes a physical action to occur at a wellsite.

18 . The non-transitory computer-readable medium of claim 13 , wherein the training comprises or generates the modified objective function, and wherein the modified objective function is based upon:

a Fourier transform that maps at least one of the first dataset or the second dataset from an original domain to a frequency domain;

an L1 norm for outliers of the first dataset and the second dataset in the frequency domain; and

first and second weighting parameters for the L1 norm.

19 . The non-transitory computer-readable medium of claim 18 , wherein the modified objective function further comprises an adversarial loss component that ensures that a generator produces generated data that matches an overall distribution of real data.

20 . The non-transitory computer-readable medium of claim 19 , wherein the modified objective function further comprises:

a frequency-domain misfit function that enforces similarity between the real data and the generated data in the frequency domain, thereby capturing structural differences that are not apparent in a time and a spatial domain; and

a data-domain misfit function that ensures that samples of the generated data resemble the real data in the original domain, thereby adding robustness to the outliers and to noise.