Feature selection for reservoir characterisation by Bayesian network

Abstract

The more accurate feature identification, the more precise reservoir characterisation. Porosity, permeability and other rock properties could be estimated and classified by analytical and intelligent methods. Feature selection plays a vital role in the process of identification. In this work, two goals are followed: first, developing Bayesian Network, K2 algorithm, as a complementary means (not an alternative) to find interrelationships of petrophysical parameters and second, feature conditioning for estimating porosity and permeability, vug and fracture detection, and net pay determination. Due to the results, bulk density log is introduced as the most important feature for characterizing the reservoir because it is found useful for identifying all the studied reservoir features.