Comparação de técnicas de inteligência computacional para classificação de dados petrográficos

Abstract

Predictive models of heterogeneities distribution and quality in hydrocarbon reservoirs are of fundamental importance for exploration and production optimization of oil and gas fields. The heterogeneities are determined by the different sedimentary petrofacies, a set of petrographic characteristics that specifies a group of rocks. The identification and classification of petrofacies is usually a time consuming procedure, and the use of computational methods can reduce the time and effort spent in the analysis. Recently, techniques derived from the computational intelligence research area have been used to assist in making decisions experts in several problems in Geosciences. The purpose of this dissertation is evaluating the performance of different techniques based on computational intelligence to predict the classification of petrographic samples belonging to the same sedimentary basin. A computational framework was developed to classify petrofacies according to their constituents. The data was collected from three different sources. The first database is formed by thin sections of Tibagi region (PR). The second by thin sections of Dom Aquino region (MS). Such thin sections are for a lithostratigraphic unit formalized in the Paraná Basin as Member Tibagi. The third by thin sections from the two previous databases. The fourth database is a set of thin sections from Mucuri member of the Espírito Santo sedimentary basin. The proposed method involves the use of classifiers, cross validation, dimensionality reduction, feature selection and the use of ensemble of constituents. The parameters involved in adjusting methods were determined by an exhaustive search procedure with cross-validation and classification metrics were used to evaluate and compare the results. The presented methodology evaluates the performance of several computational intelligence techniques, and arises as an alternative to assist the geologist in the determination and characterization of petrofacies, helping to reduce the effort in the process of individualization.