Identificação de litofácies de poços de petróleo utilizando um método baseado em redes neurais artificiais.

Abstract

The main objective of this work is to propose, implement and evaluate a method to auto matically identify lictofacies (lithological units) from well log and core data of an oil field. This is important since it can help determine whether a well is economically viable or not. A typical well log contains rocks sedimentary information occurring along a wide depth range using a resolution of under a meter, beyond porosity and permeability informations. Manu al lictofacies identification from well logs is usually time consuming, involves the analysis of large amounts of data and relies upon very specific (sometimes heuristic) knowledge. A detailed description of the lithological units can be obtained by a core sample analysis, but this is a very expensive process and is made available just to a few wells. Thus, the need of a computational method to solve the above problem becomes obvious. Our method consists of using a neural network approach to perform knowledge acquisition from a database of well logs and core data. The database was provided by the Brazilian Oil Agency (ANP) and contains data from the Namorado oil field in Rio de Janeiro. A previous attempt to solve this problem using neural networks used data from only 5 wells. In this work, we use data from 8 wells. The main modules of the proposed method were implemented and validated from a real data set. The average identification rate was around 80 %. A solution to the problem was only possible after the incorporation of a lictofacies grouping strategies and after dealing with some problematic patterns (regions of uncertain knowledge within the training and test sets).