The Ohio Geological Society

 

Thursday, April 20^th, 2023

 

1:00 PM (EST)

 

 Zoom WEBINAR

(Thanks to Eastern Section AAPG!!!!)

Zoom Registration Link:

https://us06web.zoom.us/webinar/register/WN_BtOs_ViWR6-6pmgEmy7Atg

 

 

An Introduction to Popular Machine Learning Tools for Seismic Interpreters

 

Carrie Laudon, Senior Geophysical Consultant

Geophysical Insights

 

Machine Learning and AI are hot topics in most industries, and the subsurface is no exception. 3D seismic data and seismic interpreters have been working with “Big Data” for decades. However, computing capability has finally caught up with the algorithms so that machine learning techniques can be deployed on seismic data. Companies have developed automation in seismic processing, formation tops picking, and log estimation, which utilize ML. This presentation will focus on two commercial technologies, one supervised and one unsupervised, that aid specifically in interpreting post-stack seismic data, automatic fault prediction, and multi-attribute analysis via Self Organizing Maps (SOM).

Automated Fault prediction is a supervised machine learning technique that uses convolutional neural networks to train an ML engine to detect whether samples are faults or not. This is an excellent task to automate, as fault interpretation can be quite manual and laborious. The engines can be trained by manual interpretation or by using synthetic fault data. There is evidence that synthetic engines are superior in complexity faulted terrains where interpretation out of the fault planes is a big challenge. Experience has shown that pre and post-processing steps are critical to achieving acceptable results compared to hand-picked faults. Results will be shown on a complexly faulted data set from Australia.

Unsupervised classification uses the data itself to find patterns which are called natural clusters in attribute space. SOM is used along with Principal Component Analysis (PCA) as a multi-attribute technology that allows interpreters to use machine learning techniques to quickly and effectively analyze dozens of seismic attributes and develop an improved understanding of the subsurface. An advantage over supervised classification is that SOM can find patterns in data that might not necessarily exist in the training data used in a supervised classification. Two examples will be presented, a thin bed analysis of an unconventional reservoir in the DJ Basin and a study that examined potential karst features as conduits for gas migration into the Utica/Pt Pleasant Formations in SW Pennsylvania.

Carrie Laudon Bio

Carolan (“Carrie”) Laudon has worked as a Senior Geoscience Consultant with Geophysical Insights since 2017, applying the Paradise® AI workbench and other interpretation tools for client projects. Her prior roles include Vice President of Consulting Services and Microseismic Technology for Global Geophysical Services and 17 years with Schlumberger in technical, management, and sales. Carolan Laudon holds a Ph.D. in Geophysics from the University of Minnesota and a B.S. in Geology from the University of Wisconsin Eau Claire.