Machine Learning-Assisted Petrophysical Rock Type Classification and Permeability Estimation with Flow Zone Indicators

This is the authors' abstract. We don't add key points for this paper.

The main objective of this study is to derive petrophysical characteristics and predict petrophysical rock types (PRT) from a well log from the Volve Field Dataset, using unsupervised and supervised machine learning algorithms. We utilized available core data of the reservoir with an unsupervised K-means algorithm and supervised random forest classifier to predict rock types and permeability of the well. This paper proposes a methodology that takes advantage of calculated flow zone indicators (FZI) and hydraulic units for enhanced characterization. The core data set containing 557 core plug samples, which consists of porosity, permeability, grain density, and water saturation measurements, was classified into PRTs based on core FZI and hydraulic units (HU). These calculated FZI and HU values of the core data acted as the training data set to estimate HUs for the well log using random forest classification. Permeability was then estimated from a linear regression associated with each HU, based on the calibration to calculated core FZI values. This paper documents the ability to conduct petrophysical characterization with minimal initial core data. By applying the K-means algorithm to intuitively identify sedimentary and diagenetic characteristics of the reservoir, and by applying the random forest algorithm on a conventional parameter such as FZI, one can sufficiently estimate petrophysical properties and classify PRTs from well logs in a time-saving manner.