--> Wireline Log Data Conditioning to Improve Machine Learning Based Facies Identification Across the Well

AAPG Asia Pacific Technical Symposium

Datapages, Inc.Print this page

Wireline Log Data Conditioning to Improve Machine Learning Based Facies Identification Across the Well

Abstract

Wellbore Core is the most important data for geologist in identifying facies in subsurface realm, unfortunately core is not always available. In contrast with wellbore core, wireline log is almost always available in wellbore, thus geologist develop methods to identify depositional facies. The technique that being developed are more into assessing the data qualitatively, they use log vertical pattern for their interpretation. For most of the time the results of the interpretation are rather subjective, meaning every geologist has different eye in making facies interpretation, even the same geologist might have different interpretation on the second time they interpreting the same data. It even getting worst when interpreting different well that have different log statistical characteristic. Therefore, a more consistent approach is needed to reduce interpretation subjectivity. Wireline log is stack of formatted data that contains many geological information, such as rock property as well as facies. With representative sample, machine learning capable to identify facies from well log data. In previous research, single well facies identification has satisfying result, however when it comes with different well, especially that have log statistical characteristic facies identification result are not very satisfying. There are three methods to improve machine learning based facies identification accuracy: more facies sample using wireline log data, tuning machine learning algorithm, and lastly data conditioning. This research will focus on the effect of data conditioning in Machine Learning Based Facies Identification. The conditioning that will be discussed in this research are statistical correction, mathematical operation, increasing (or reducing) data dimensionality, and log resolution reduction.