About This Item

Share This Item

The AAPG/Datapages Combined Publications Database

AAPG Bulletin

Abstract

DOI: 10.1306/12162121016

Machine learning applications to seismic structural interpretation: Philosophy, progress, pitfalls, and potential

Kellen L. Gunderson,1 Zhao Zhang,2 Barton Payne,3 Shuxing Cheng,4 Ziyu Jiang,5 and Atlas Wang6

1Chevron Technical Center, Houston, Texas; present address: Zanskar Geothermal & Minerals, Inc., Provo, Utah; [email protected]
2Chevron Technical Center, Houston, Texas; [email protected]
3Chevron Technical Center, Houston, Texas; [email protected]
4Chevron Technical Center, Houston, Texas; [email protected]
5Texas A&M University, College Station, Texas; [email protected]
6Texas A&M University, College Station, Texas; present address: The University of Texas at Austin, Austin, Texas; [email protected]

ABSTRACT

Seismic structural interpretation is a discipline of structural geology that uses seismic reflection data to interpret geologic structures in the subsurface. The advancement of machine learning technology, especially in image analysis methods like computer vision, has opened new frontiers for the discipline. Here, we review the current state of machine learning applied to seismic structural interpretation, outlining the philosophy, progress, pitfalls, and potential of the technology when applied to the characterization of subsurface structures. Seismic structural interpretation incorporates seismic image analysis, analogue use, and structural modeling to interpret geologic structures in the subsurface; machine learning approaches seek to supplement these current methods with workflows that benefit from a computer’s ability to continually improve with more data. Most of the current research in the discipline is focused on using convolutional neural networks (CNNs), a computer vision approach, to interpret faults, salt bodies, and structural traps. Automated structural trap interpretation presents unique challenges to the CNN approach, which we propose are best addressed using a geometric invariance-enforced deep learning method presented here. The CNNs and other computer vision methods offer performance advantages to seismic attribute-based approaches to structural interpretation, but still suffer from the potential pitfalls of image-only interpretation, like lacking geologic context. Additional challenges remain around constructing global, open training data, and developing new artificial intelligence approaches that can incorporate kinematic or geomechanical structural models in seismic interpretation workflows.

Pay-Per-View Purchase Options

The article is available through a document delivery service. Explain these Purchase Options.

Watermarked PDF Document: $14
Open PDF Document: $24

AAPG Member?

Please login with your Member username and password.

Members of AAPG receive access to the full AAPG Bulletin Archives as part of their membership. For more information, contact the AAPG Membership Department at [email protected].