Rule Induction Algorithm for Application to Geological and Petrophysical Data

Abstract

Very large geological, geophysical, and petrophysical databases often contain multiple data types that must be interpreted for application to subsurface modeling. Significant advances in discovering complex and even nonintuitive data relationships could lead to better predictions. There is a litany of data analysis techniques used today, including cluster analysis, principal component analysis, discriminant analysis, parametric and nonparametric regression, and N-dimensional histograms. Regression techniques and neural networks have in common their multivariate combination of predictor variables. These techniques may be good at interpolating within the data boundaries of the training data, but may be poor for extrapolation because of the lack of understanding of the underlying relationships in the variables. Alternatively, machine learning and data mining technologies including Rough Sets hold the promise of finding data category relationships and expressing those in a rule-based language. This paper presents a novel rule induction algorithm derived from these machine-learning techniques, developed for reservoir characterization with geological and geophysical data. A set of facies models with systematical changing in the geometric features is synthesized. The geometric features are coded and the effective permeability is calculated. Rules between effective permeability and geometric features are deducted by using the proposed technique. The consistence of the deducted rules with those implemented in the data synthesization exhibit the effectivity of the proposed technique. Further a second example of facies assignment from wireline logs is used to test the proposed technique. The deducted rules are confirmed by the geologists who spend significant time trying to summarize rules from the well logs. The probability feature of the rules and the distingushibility analysis feature of the proposed technique supplied additional information for the geologist to reconsider their original distinction among facies.