Random noise attenuation for seismic data

A system comprising:

• a processor;
• and
• a memory having processor-readable instructions stored therein, which, when executed by the processor, cause the processor to perform a plurality of functions, including functions to:
• train, based on a random noise library and on a clean data sample of additional seismic data, a first model to extract random noise from seismic datasets, wherein the random noise library comprises a plurality of different noise profiles, wherein the clean data sample lacks random noise;
• train a second model to reconstruct signals leaked into the random noise extracted by the first model;
• obtain a first seismic dataset corresponding to a first subsurface reservoir formation, wherein the first seismic dataset includes random noise;
• extract, using the trained first model, at least a portion of the random noise from the first seismic dataset, wherein the portion of the random noise includes a signal corresponding to a portion of the first seismic dataset;
• reconstruct, using the trained second model, the signal corresponding to the portion of the first seismic dataset included with the portion of the random noise extracted by the trained first model;
• generate a cleaned seismic dataset based on the first seismic dataset and the signal reconstructed from the portion of the random noise extracted from the first seismic dataset and based on excluding the portion of the random noise extracted from the first seismic dataset, wherein the cleaned seismic dataset includes a quantity of random noise that is less than that included in the first seismic dataset;
• output, to a display in communication with the processor, the cleaned seismic dataset;
• and
• perform one or more downhole operations comprising at least one of a drilling operation, a well completion operation, or an injection stimulation operation for recovering hydrocarbons in the first subsurface reservoir formation based on the cleaned seismic dataset.

A non-transitory computer-readable storage medium comprising computer-readable instructions stored therein, which, when executed by a computer, cause the computer to perform a plurality of functions, including functions to:

• train, based on a random noise library and on a clean data sample of additional seismic data, a first model to extract random noise from seismic datasets, wherein the random noise library comprises a plurality of different noise profiles, wherein the clean data sample lacks random noise;
• train a second model to reconstruct signals leaked into the random noise extracted by the first model;
• obtain a first seismic dataset corresponding to a first subsurface reservoir formation, wherein the first seismic dataset includes random noise;
• extract, using the trained first model, at least a portion of the random noise from the first seismic dataset, wherein the portion of the random noise includes a signal corresponding to a portion of the first seismic dataset;
• reconstruct, using the trained second model, the signal corresponding to the portion of the first seismic dataset included with the portion of the random noise extracted by the trained first model;
• generate a cleaned seismic dataset based on the first seismic dataset and the signal reconstructed from the portion of the random noise extracted from the first seismic dataset and based on excluding the random noise extracted from the first seismic dataset, wherein the cleaned seismic dataset includes a quantity of random noise that is less than that included in the first seismic dataset;
• output, to a display in communication with the computer, the cleaned seismic dataset;
• and
• perform one or more downhole operations comprising at least one of a drilling operation, a well completion operation, or an injection stimulation operation for recovering hydrocarbons in the first subsurface reservoir formation based on the cleaned seismic dataset.