[First Hit]

2019 AAPG Annual Convention and Exhibition:

Datapages, Inc.Print this page

Comparison of Deep Learning Fault Previous HitInterpretationNext Hit From Seismic Data With Traditional and Attribute Based Techniques

Abstract

The growth in artificial intelligence, and Deep Learning in particular, now gives us the potential for a much faster and more accurate fault delineation than has ever been possible before. By combining objective compute power with human cognition, Deep Learning algorithms now give us a much cleaner and more accurate fault interpretation.This paper presents a comparative study of the results of a traditional manual fault Previous HitinterpretationNext Hit, an attribute driven fault Previous HitinterpretationNext Hit, and the results from a Deep Learning algorithm. The results show that the Deep Learning algorithm significantly outperforms attribute analysis in terms of the accuracy of the fault detection and the almost complete lack of background noise. It also outperforms manual Previous HitinterpretationNext Hit in terms of the speed of Previous HitinterpretationNext Hit whilst maintaining accuracy. One of the most striking observations of this study is the ability of the Deep Learning network to perform well when the data quality is poor, its ability to differentiate faults from imaging related artefacts in the data, even when the seismic expression is similar. These results show the enormous power of Deep Learning to extract a more accurate fault Previous HitinterpretationNext Hit in less time than either a manual Previous HitinterpretationTop or attribute driven analysis.