Application of Machine Learning and Deep Learning for Complex Fault Network Characterizationon the North Slope, Alaska

Abstract

The deep subsurface geology on the North Slope, Alaska is structurally complex and pervasively fractured. Faults/fractures control fluid flow (hydrocarbon and water) through rocks. Although seismic data is generally used to “manually” identify faults in the subsurface, the extreme nature of these discontinuities, such as their numbers per square miles and complexities, do not allow geoscientists to fully understand the multi-phase fault development history of northern Alaska over geologic time. Physics-based seismic attributes such as 3D curvature can be used to detect faults. However, this process is computationally intensive and requires a detailed understanding of the limits of the attributes in specific geologic settings and the quality of the data. In this study, a large 3D seismic survey over an area of 270 square miles on the North Slope was used for detailed fault characterization. Curvature attribute-assisted horizon mapping revealed the presence of three major extensional fault network along NW-SE, N-S, and E-W directions, many of which affect multiple source and reservoirs such as the Shublik, Sag River, and Kuparuk formations. Next, Multi-layer Perceptron Neural Network (MLPNN) and Convolutional Neural Network (CNN) were used to classify and predict faults in the seismic survey automatically. The results show that both MLPNN and CNN can be used for fault classification with high accuracy (>85%) in limited time; however, CNN-based fault classification does not require any seismic attributes as input to the neural network model as opposed to the MLPNN. The original seismic data with labeled faults can be directly used in the CNN model for automated fault classification, thereby, bypassing the traditional way of seismic attribute-assisted fault detection.

AAPG Datapages/Search and Discovery Article #90339 ©2019 AAPG Pacific Section Convention 2019, Long Beach, California, April 1-3, 2019