Machine Learning for Simple Petrophysical Analysis

Abstract

The Bureau of Ocean Energy Management (BOEM) has initiated machine learning (ML) tests in collaboration with the California State University Northridge’s Center for Geospatial Science and Technology (CGST) using small data sets from wells on the Southern California Outer Continental Shelf. Work is in progress to create larger and more accessible data sets as further ML testing is conducted. This talk will describe what has been accomplished so far and explain the short term goals to exploit this new technology. BOEM’s first ML test was able to distinguish between sands and claystones using a decision tree classifier trained with about 100 sidewall core descriptions and wireline logs. However, the accuracy was only about 75%. Subsequently CGST used a larger dataset to see which ML algorithm was most effective in distinguishing between sands and claystones. The best results, with an accuracy of about 80%, were obtained by a gradient-boosting classifier in single-well tests. Current work is focused on making a larger analytical dataset that can be filtered spatially and by geologic formation. The intention is to continue testing ML applications for lithological classification and permeability estimation.

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