Towards a Basin-Scale Lithofacies Model - An Integrated Machine Learning Approach Using
Well
Logs and Core for the Permian Basin
Abstract
Understanding depositional environments and lithologies are key steps in developing a comprehensive understanding of unconventional reservoirs. However, to characterize reservoirs on a basin-scale, there lies a need to efficiently analyze and integrate large dataset. This time consuming challenge is compounded by significant heterogeneities that exist within a basin and issues associated with data integrity. With internal tools to perform formation top propagation and well
log
correction, we are able to accelerate the production of clean, high-density dataset prior to
analysis
, which has helped to significantly reduce uncertainty and enhance our traditional geoscience workflows. In this study, we present an integrated workflow that allows us to better understand
log
-based depositional environments, which are then used to construct a predictive model for facies classification at the basin-scale. This workflow employs unsupervised learning and supervised learning techniques to rapidly analyze
log
and core data. Firstly, cluster
analysis
of 5000 raw
well
logs using both data reduction and data normalization techniques has allowed for the
analysis
of each formation within the basin, enabling the interpretation of depositional environments at the basin scale. To further capture the varying degrees of heterogeneity observed within each of the target formations, clustering algorithms for each formation were chosen based on key performance indicators, and a lithofacies scheme was established from cluster
analysis
of core data from each of the key formations. Secondly, a predictive facies classification was developed using
well
logs and the cluster-derived lithofacies scheme. In order to incorporate multiple scales and to meet the expert-defined quality threshold required by our geologists and petrophysicists, the modelling of facies was restricted to
log
-derived clusters. Benchmarking results against historical regional studies indicate that this workflow can very quickly establish high-quality reservoir characteristics at a basin-scale.
AAPG Datapages/Search and Discovery Article #90350 © 2019 AAPG Annual Convention and Exhibition, San Antonio, Texas, May 19-22, 2019