Machine learning training based on dual loss functions
A computer-implemented method for seismic processing includes receiving a seismic training input image, generating, using a first portion of a machine learning model, a first output based at least in part on the seismic training input image, generating, using a second portion of the machine learning model, a second output based at least in part on the seismic training input image, generating a loss function based at least in part on comparing at least two of the first output, a deterministic first label synthetically generated and representing a deterministic ground truth for the first output, the second output, and a non-deterministic second label representing a non-deterministic ground truth for the second output, and refining the first portion, the second portion, or both of the machine learning model based at least in part on the loss function.
1 . A method for generating a model of a subterranean formation and controlling drilling operations based on the model, the method comprising:
receiving a seismic training input image;
generating, using a first portion of a machine learning model, a first output based at least in part on the seismic training input image;
generating, using a second portion of the machine learning model, a second output based at least in part on the seismic training input image;
generating a loss function based at least in part on comparing at least two of:
the first output;
a deterministic first label synthetically generated and representing a deterministic ground truth for the first output;
the second output; and
a non-deterministic second label representing a non-deterministic ground truth for the second output;
refining the first portion, the second portion, or both of the machine learning model based at least in part on the loss function by:
training the machine learning model based on deterministic first labels and non-deterministic second labels;
applying the seismic training input image to a deterministic interpretation section to obtain a deterministic interpretation output;
applying the training input to a non-interpretation section to obtain a non-deterministic output; and
correcting errors in the non-deterministic output on subsequent iterations using the loss function;
generating, based on the refined first portion and the refined second portion, a plurality of geological interpretation options representing alternative models of the subterranean formation;
ranking the plurality of geological interpretation options using scores determined from outputs of the refined portions;
selecting, from among the plurality of geological interpretation options, a selected geological interpretation option;
determining, based on the selected geological interpretation option, a drilling action comprising at least one mitigating action associated with the selected geological interpretation option; and
transmitting a signal to a controller to cause a drilling tool to perform the drilling action, the signal being at least one of generated or transmitted by the refined first portion, the refined second portion, or both of the machine learning model, the drilling action comprising modifying a drilling direction toward hydrocarbons or away from obstructions identified in the selected geological interpretation option.
2 . The method of claim 1 , wherein:
the first portion is a deterministic portion; and
the second portion is a non-deterministic portion.
3 . The method of claim 2 , wherein:
the first output is a deterministic output; and
the second portion is a non-deterministic output.
4 . The method of claim 3 , further comprising generating the deterministic first label by applying the seismic training input image to a physical model.
5 . The method of claim 1 , wherein the refining comprises adjusting a weight assigned to the first output, the deterministic first label, the second output, the non-deterministic second label, or a combination thereof based at least in part on the loss function to reduce an error of the first portion of the machine learning model, the second portion of the machine learning model, or both.
6 . The method of claim 1 , wherein the refining the first portion of the machine learning model is based on the first output and the non-deterministic second label, and not on the second output and the deterministic first label.
7 . The method of claim 1 , wherein the refining the second portion of the machine learning model is based on the second output and the deterministic first label, and not on the first output and the non-deterministic second label.
8 . The method of claim 1 , further comprising identifying hydrocarbons in the seismic training input image or a seismic operational input image using the refined machine learning model.
9 . A computing system, comprising:
one or more processors; and
a memory system comprising 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 comprising:
receiving a seismic training input image;
generating, using a first portion of a machine learning model of a subterranean formation, a first output based at least in part on the seismic training input image;
generating, using a second portion of the machine learning model of the subterranean formation, a second output based at least in part on the seismic training input image;
generating a loss function based at least in part on comparing at least two of:
the first output;
a deterministic first label synthetically generated and representing a deterministic ground truth for the first output;
the second output; and
a non-deterministic second label representing a non-deterministic ground truth for the second output;
refining the first portion, the second portion, or both of the machine learning model based at least in part on the loss function by:
training the machine learning model based on deterministic first labels and non-deterministic second labels;
applying the seismic training input image to a deterministic interpretation section to obtain a deterministic interpretation output;
applying the training input to a non-interpretation section to obtain a non-deterministic output; and
correcting errors in the non-deterministic output on subsequent iterations using the loss function;
generating, based on the refined first portion and the refined second portion, a plurality of geological interpretation options representing alternative models of the subterranean formation;
ranking the plurality of geological interpretation options using scores determined from outputs of the refined portions;
selecting, from among the plurality of geological interpretation options, a selected geological interpretation option;
determining, based on the selected geological interpretation option, a drilling action comprising at least one mitigating action associated with the selected geological interpretation option; and
transmitting a signal to a controller to cause a drilling tool to perform the drilling action, the signal being at least one of generated or transmitted by the refined first portion, the refined second portion, or both of the machine learning model, the drilling action comprising modifying a drilling direction toward hydrocarbons or away from obstructions identified in the selected geological interpretation option.
10 . The computing system of claim 9 , wherein:
the first portion is a deterministic portion; and
the second portion is a non-deterministic portion.
11 . The computing system of claim 10 , wherein:
the first output is a deterministic output; and
the second portion is a non-deterministic output.
12 . The computing system of claim 11 , wherein the deterministic first label is generated by applying the seismic training input image to a physical model.
13 . The computing system of claim 9 , wherein the refining comprises adjusting a weight assigned to the first output, the deterministic first label, the second output, the non-deterministic second label, or a combination thereof based at least in part on the loss function to reduce an error of the first portion of the machine learning model, the second portion of the machine learning model, or both.
14 . The computing system of claim 9 , wherein the refining the first portion of the machine learning model is based on the first output and the non-deterministic second label, and not on the second output and the deterministic first label.
15 . The computing system of claim 9 , wherein the refining the second portion of the machine learning model is based on the second output and the deterministic first label, and not on the first output and the non-deterministic second label.
16 . The computing system of claim 9 , wherein the operations further comprise:
identifying hydrocarbons in a seismic operational input image using the refined machine learning model; and
transmitting a signal to wellsite equipment based at least in part upon a location of the hydrocarbons in the seismic operational input image.
17 . A computing system, comprising:
one or more processors; and
a memory system comprising 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 comprising:
receiving a seismic training input image;
generating, using a first portion of a machine learning model of a subterranean formation, a first output based at least in part on the seismic training input image;
generating, using a second portion of the machine learning model of the subterranean formation, a second output based at least in part on the seismic training input image;
generating a loss function based at least in part on comparing:
the first output;
a deterministic first label synthetically generated and representing a deterministic ground truth for the first output;
the second output; and
a non-deterministic second label representing a non-deterministic ground truth for the second output;
refining the first portion and the second portion of the machine learning model based at least in part on the loss function by:
training the machine learning model based on deterministic first labels and non-deterministic second labels;
applying the seismic training input image to a deterministic interpretation section to obtain a deterministic interpretation output;
applying the training input to a non-interpretation section to obtain a non-deterministic output; and
correcting errors in the non-deterministic output on subsequent iterations using the loss function;
generating, based on the refined first portion and the refined second portion, a plurality of geological interpretation options representing alternative models of the subterranean formation;
ranking the plurality of geological interpretation options using scores determined from outputs of the refined portions;
selecting, from among the plurality of geological interpretation options, a selected geological interpretation option;
determining, based on the selected geological interpretation option, a drilling action comprising at least one mitigating action associated with the selected geological interpretation option; and
transmitting a signal to a controller to cause a drilling tool to perform the drilling action, the signal being at least one of generated or transmitted by the refined first portion, the refined second portion, or both of the machine learning model, the drilling action comprising modifying a drilling direction toward hydrocarbons or away from obstructions identified in the selected geological interpretation option.
18 . The computing system of claim 17 , wherein:
the first portion is a deterministic portion; and
the second portion is a non-deterministic portion.
19 . The computing system of claim 18 , wherein:
the first output is a deterministic output; and
the second portion is a non-deterministic output.
20 . The computing system of claim 17 , wherein:
the operations further comprise:
identifying hydrocarbons in a seismic operational input image using the refined machine learning model; and
transmitting a signal to wellsite equipment based at least in part upon a location of the hydrocarbons in the seismic operational input image; and
the signal instructs the wellsite equipment to drill toward the hydrocarbons in a subterranean formation.
21 . The method of claim 1 , further comprising:
receiving a seismic operational input image; and
updating the selected geological interpretation option based at least in part on the seismic operational input image.
22 . The method of claim 1 , wherein:
the refining the first portion comprises adjusting a weight corresponding to a comparison between the first output and the non-deterministic second label; and
the refining the second portion comprises adjusting a weight corresponding to a comparison between the second output and the deterministic first label.
23 . The method of claim 1 , wherein:
the refining the first portion and the second portion of the machine learning model based at least in part on the loss function comprises adjusting weights within a deterministic interpretation section and a non-deterministic interpretation section to reduce errors determined using a deterministic label and a non-deterministic label that is manually generated by a subject-matter expert; and
the method further comprises:
after the training, using the machine learning model to output a deterministic interpretation and a non-deterministic interpretation of a seismic operational input image;
using a feedback loop executed on an algorithmic basis at a computing device to iteratively evaluate the deterministic interpretation and the non-deterministic interpretation and to select one of the deterministic interpretation and the non-deterministic interpretation as a geological interpretation of a subterranean formation;
identifying hydrocarbons in the seismic operational input image using the selected geological interpretation; and
transmitting a signal to wellsite equipment based at least in part upon a location of the hydrocarbons in the seismic operational input image to instruct the wellsite equipment to drill toward the hydrocarbons or away from obstructions in the subterranean formation.