Fault interpretation and feature learning on full azimuth stacks

A system, comprising:

• a data transceiver, configured to receive input parameters and a set of stacked azimuthal data for a subterranean formation, wherein the set of stacked azimuthal data includes at least two azimuthal data, and the set of stacked azimuthal data represents at least two angles of seismic detection;
• a machine learning system, configured to execute one or more ensemble azimuthal algorithms using the input parameters and the azimuthal data;
• and
• a subterranean feature processor, configured to communicate with the data transceiver and the machine learning system, utilize the machine learning system to generate a feature result, wherein a set of feature probability maps is generated, where each feature probability map is the set of feature probability maps represents a probability of a subterranean formation feature corresponding to one stacked azimuthal data in the set of stacked azimuthal data, each probability map in the set of feature probability maps is grouped with a respective stacked azimuthal data in the set of stacked azimuthal data where each group forms an ensemble azimuthal dataset, the machine learning system includes more than one machine learning algorithm, each machine learning algorithm in the more than one machine learning algorithm utilizes a different subterranean formation model, and each machine learning algorithm is optimized using a learning estimator using each ensemble azimuthal dataset, and a result from each of machine learning algorithms is combined to compute the feature result, where the feature result represents an approximate location, an approximate size, and an approximate orientation of the subterranean formation feature, and the feature result is used to modify a drilling parameter to change a direction of a drilling assembly located downhole a borehole, where each subterranean formation model utilizes a combination that is different from other subterranean formation models, including an operating learning algorithm, a weighting matrix to determine how each layer in the set of stacked azimuthal data is combined, and a weighting algorithm to determine how to combine the feature probability maps in the set of feature probability maps.

A computer program product having a series of operating instructions stored on a non-transitory computer-readable medium that directs a data processing apparatus when executed thereby to perform operations to determine a subterranean formation feature within a subterranean formation, the operations comprising:

• receiving a set of stacked azimuthal data for a subterranean formation, wherein the set of stacked azimuthal data represents two or more angles of seismic detection;
• generating a set of feature probability maps, where each feature probability map in the set of feature probability maps represents a probability of a subterranean formation feature corresponding to one stacked azimuthal data in the set of stacked azimuthal data;
• grouping each feature probability map in the set of feature probability maps with a respective stacked azimuthal data in the set of stacked azimuthal data, where each group forms an ensemble azimuthal dataset in a set of ensemble azimuthal datasets;
• optimizing more than one machine learning estimator using each of the ensemble azimuthal datasets;
• combining a result from more than one machine learning estimator to compute a feature result, wherein each machine learning estimator in the more than one machine learning estimator utilizes a different subterranean formation model, an input to each machine learning estimator are the set of ensemble azimuthal datasets, and the feature result represents an approximate location, an approximate size, and an approximate orientation of the subterranean formation feature, where each subterranean formation model utilizes a combination that is different from other subterranean formation models, including an operating learning algorithm, a weighting matrix to determine how each layer in the set of stacked azimuthal data is combined, and a weighting algorithm to determine how to combine the feature probability maps in the set of feature probability maps;
• and
• modifying a drilling parameter to change a direction of a drilling assembly located downhole a borehole using the feature results.