Improving Seismic Interpretation Efficiency and Accuracy Using Supervised Machine Learning to Optimize Fault Interpretation Workflows

Abstract

Interpreting faults in seismic data is an important step when constructing structural models of the subsurface. Many seismic attributes have been developed that are sensitive to fault-induced discontinuities and can be used to build fault models relatively quickly. However, parameterizing automated fault interpretation algorithms such that they detect faults with the same accuracy as an expert human interpreter remains a challenge because there are usually many interdependent parameters in these workflows requiring tuning, and the optimal parameter set is strongly dataset dependent. Consequently, time spent optimizing fault attribute and extraction parameters can offset the efficiencies gained by using an automated interpretation method over manual fault-picking approaches.

This paper discusses a hybrid fault interpretation technique that uses a small set of manually interpreted faults to condition the parameter tuning of an automated fault interpretation workflow. The approach combines a machine-learning based hyper-parameter optimization technique, a best-of-class fault attribute (Hale, 2013), and domain-expert input in the form of labelled fault images. Under the proposed scheme, an interpreter needs only to manually pick faults on a few lines of a 3D seismic dataset. These interpreted sections are then used to define an objective function that, in turn, allows automatic optimization of the fault interpretation workflow parameter. The parameter space is intelligently searched to identify the parameter set whose fault prediction best matches the input expert-labelled data. Once the optimization process has converged, the optimal workflow parameters are used to compute a high-quality fault interpretation for the entire seismic volume. Using the hybrid fault interpretation workflow, the time taken to obtain reliable results from automatic fault interpretation algorithms can be reduced, thus helping improve the overall efficiency of seismic interpretation workflows and freeing geoscientists to perform more advanced analyses to inform better reservoir decisions.

References:

Hale, Dave (2013). “Methods to compute fault images, extract fault surfaces, and estimate fault throws from 3D seismic images.” GEOPHYSICS, 78(2), O33–O43. DOI: 10.1190/GEO2012-0331.1.

AAPG Datapages/Search and Discovery Article #90323 ©2018 AAPG Annual Convention and Exhibition, Salt Lake City, Utah, May 20-23, 2018