Method for predicting a geophysical model of a subterranean region of interest
A system and methods are disclosed for determining a model of a subterranean region. The method includes obtaining an observed dataset and a current model for the subterranean region, simulating a dataset from the current model, and determining a data penalty function based on a difference between the observed and simulated datasets. The method further includes training a machine learning (ML) network to predict a model from the observed dataset and determining the predicted model using the trained ML network. The method further includes determining a first model penalty function based on the current model, a second model penalty function based on a difference between the current and the predicted models, and a composite penalty function based on a weighted sum of the data penalty function, the first and the second model penalty functions. Finally, the method includes determining the model based on an extremum of a composite penalty function.
1 . A method for determining a geophysical model of a subterranean region, comprising:
obtaining an observed geophysical dataset from the subterranean region;
obtaining a current geophysical model for the subterranean region;
executing, by a computer processor, a training procedure that iteratively updates parameters of a deep learning neural network using a plurality of geophysical training models and a corresponding geophysical training dataset for each geophysical training model to produce a trained machine learning (ML) network configured to generate a predicted geophysical model;
generating, by the trained ML network executed on the computer processor, the predicted geophysical model from the observed geophysical dataset;
iteratively, until a convergence criterion is satisfied, performing steps comprising:
simulating, by a forward modeling method executed on the computer processor, a simulated geophysical dataset from the current geophysical model;
generating, using the computer processor, a data penalty function based on a difference between the observed geophysical dataset and the simulated geophysical dataset;
generating, using the computer processor, a first model penalty function based on the current geophysical model;
generating, using the computer processor, a second model penalty function based on a difference between the current geophysical model and the predicted geophysical model;
generating, using the computer processor, a composite penalty function based, at least in part, on a weighted sum of the data penalty function, the first model penalty function, and the second model penalty function;
generating, using the computer processor, an updated geophysical model based, at least in part, on finding an extremum of the composite penalty function; and
replacing the current geophysical model with the updated geophysical model;
wherein the geophysical model of the subterranean region is the updated geophysical model obtained upon satisfaction of the convergence criterion.
2 . The method of claim 1 , further comprising:
determining a location of a hydrocarbon reservoir based, at least in part, on the geophysical model; and
planning, using a wellbore planning system, a planned wellbore path to intersect the hydrocarbon reservoir.
3 . The method of claim 2 , further comprising drilling a wellbore guided by the planned wellbore path using a drilling system.
4 . The method of claim 1 , wherein executing the training procedure comprises:
determining the plurality of geophysical training models and the corresponding geophysical training dataset for each geophysical training model based, at least in part, on the current geophysical model; and
training the ML network based, at least in part, on finding an extremum of a training penalty function measuring a difference between each geophysical training model and a corresponding predicted geophysical training model.
5 . The method of claim 4 , wherein executing the training procedure further comprises:
augmenting the plurality of geophysical training models and the corresponding geophysical training datasets with the current geophysical model and the simulated geophysical dataset for the current geophysical model; and
retraining the trained ML network based, at least in part, on the augmented plurality of geophysical training models and the corresponding geophysical training datasets.
6 . The method of claim 1 , wherein the current geophysical model and the training geophysical models each comprises a plurality of geophysical model types selected from a seismic model, a density model, and a resistivity model.
7 . The method of claim 6 , wherein the first model penalty function comprises:
a measure of a first difference between the current geophysical model and a reference geophysical model;
a measure of a second difference between a spatial variation of each of the geophysical model types; and
a measure of a first correlation between values of each of the geophysical model types.
8 . The method of claim 1 , wherein the second model penalty function comprises:
a measure of a third difference between the current geophysical model and the predicted geophysical model predicted by the trained ML network;
a measure of a fourth difference between a spatial variation of the current geophysical model and a spatial variation of the predicted geophysical model; and
a measure of a second correlation between values of the current geophysical model and the predicted geophysical model.
9 . The method of claim 1 , wherein the extremum comprises a minimum.
10 . The method of claim 1 , wherein the forward modeling method comprises a physics-based forward modeling method.
11 . A non-transitory computer readable medium, storing instructions executable by a computer processor, the instructions comprising functionality for:
receiving an observed geophysical dataset from a subterranean region;
receiving a current geophysical model for the subterranean region;
executing a training procedure that iteratively updates parameters of a deep learning neural network using a plurality of geophysical training models and a corresponding geophysical training dataset for each geophysical training model to produce a trained machine learning (ML) network configured to generate a predicted geophysical model;
generating, by the trained ML network executed on the computer processor, the predicted geophysical model from the observed geophysical dataset;
iteratively, until a convergence criterion is satisfied, performing steps comprising
simulating, by a forward modeling method executed on the computer processor, a simulated geophysical dataset from the current geophysical model;
generating a data penalty function based on a difference between the observed geophysical dataset and the simulated geophysical dataset;
generating a first model penalty function based on the current geophysical model;
generating a second model penalty function based on a difference between the current geophysical model and the predicted geophysical model;
generating a composite penalty function based, at least in part, on a weighted sum of the data penalty function, the first model penalty function, and the second model penalty function;
generating an updated geophysical model based, at least in part, on finding an extremum of the composite penalty function; and
replacing the current geophysical model with the updated geophysical model;
wherein a geophysical model of the subterranean region is the updated geophysical model obtained upon satisfaction of the convergence criterion.
12 . The non-transitory computer readable medium of claim 11 , the instructions further comprising the functionality for:
determining a location of a hydrocarbon reservoir based, at least in part, on the geophysical model; and
planning, using a wellbore planning system, a wellbore path to intersect the hydrocarbon reservoir.
13 . The non-transitory computer readable medium of claim 11 , the instructions further comprising the functionality for:
determining the plurality of geophysical training models and the corresponding geophysical training dataset for each geophysical training model based, at least in part, on the current geophysical model of the subterranean region; and
training the ML network based, at least in part, on finding an extremum of a training penalty function measuring the difference between each geophysical training model and a corresponding predicted geophysical training model.
14 . The non-transitory computer readable medium of claim 13 , the instructions further comprising the functionality for:
augmenting the plurality of geophysical training models and the corresponding geophysical training datasets with the current geophysical model and the simulated geophysical dataset for the current geophysical model; and
retraining the trained ML network based, at least in part, on the augmented plurality of geophysical training models and the corresponding geophysical training datasets.
15 . The non-transitory computer readable medium of claim 11 , wherein the current geophysical model and the training geophysical models each comprises a plurality of geophysical model types selected from a seismic model, a density model, and a resistivity model.
16 . The non-transitory computer readable medium of claim 15 , wherein the first model penalty function comprises:
a measure of a first difference between the current geophysical model and a reference geophysical model;
a measure of a second difference between a spatial variation of each of the geophysical model types; and
a measure of a first correlation between values of each of the geophysical model types.
17 . The non-transitory computer readable medium of claim 11 , wherein the second model penalty function comprises:
a measure of a third difference between the current geophysical model and the predicted geophysical model predicted by the trained ML network;
a measure of a fourth difference between a spatial variation of current geophysical model and a spatial variation of the predicted geophysical model; and
a measure of a second correlation between values of current geophysical model and the predicted geophysical model.
18 . The non-transitory computer readable medium of claim 11 , wherein the extremum comprises a minimum.
19 . A system, comprising:
a computer system, configured to:
receive an observed geophysical dataset from a subterranean region,
receive a current geophysical model for the subterranean region,
execute a training procedure that iteratively updates parameters of a deep learning neural network using a plurality of geophysical training models and a corresponding geophysical training dataset for each geophysical training model to produce a trained machine learning (ML) network configured to generate a predicted geophysical model;
generate, by the trained ML network, the predicted geophysical model from the observed geophysical dataset;
iteratively, until a convergence criterion is satisfied, perform steps comprising:
simulate, by a forward modeling method, a simulated geophysical dataset from the current geophysical model,
generate a data penalty function based on a difference between the observed geophysical dataset and the simulated geophysical dataset,
generate a first model penalty function based on the current geophysical model,
generate a second model penalty function based on a difference between the current geophysical model and the predicted geophysical model,
generate a composite penalty function based, at least in part, on a weighted sum of the data penalty function, the first model penalty function, and the second model penalty function,
generate an updated geophysical model based, at least in part, on finding an extremum of the composite penalty function, and
replace the current geophysical model with the updated geophysical model,
wherein a geophysical model of the subterranean region is the updated geophysical model obtained upon satisfaction of the convergence criterion; and
generate a location of a hydrocarbon reservoir based, at least in part, on the geophysical model of the subterranean region; and
a wellbore planning system configured to plan a planned wellbore path to intersect the location of the hydrocarbon reservoir.
20 . The system of claim 19 , further comprising a drilling system to drill a wellbore guided by the planned wellbore path.