Abstract

This paper describes a method of advanced data processing for the inverse problem of lithofacies prediction from well logs using fuzzy partitioning systems. A fuzzy partitioning system consists of a set of fuzzy If-Then rules of the form "If bulk density (ρ_b) is low and neutron porosity (_CNL) is high Then classify pattern x=(ρ_bCNL) as Facies F_i". In this paper, we introduce an intelligent method for the problem of fuzzy rule generation based on fuzzy clustering. Fuzzy clustering is used to detect structures in the multidimensional space of the available well log readings. Each cluster detected is a potential fuzzy classification rule. By applying fuzzy validity measures an optimum number of fuzzy clusters can be found. Using this approach, the number of rules, the antecedent membership functions and other parameters that constitute the fuzzy partitioning system are derived in an automatic way.

The aim is to find a minimum set of fuzzy classification rules that can correctly classify all log training patterns. Unlike traditional methods of predicting lithofacies, this approach does not require prior knowledge about the partitioning of the well log readings or any assumption of the facies probability densities. Computer simulations using selected well log responses and facies description from a clastic and carbonate sequence in the Maracaibo Basin (western Venezuela) examine the performance of the fuzzy rule-based classification approach. The performance of the fuzzy classification method is evaluated against the facies classification results using conventional statistical analysis.