Neural Network and Logical Fuzzy application in Brazilian Carbonate formations using conventional well logging, NMR and core data for permeability prediction

Abstract

Petrophysical information, such as permeability and porosity, is of great importance for oil reserve evaluation. Rock petrophysics measurements involve some degree of uncertainty because conventional well logging and nuclear magnetic resonance data are indirect measurements. Data from core samples are more accurate, depending on the laboratory conditions, but this information belongs to a specific depth and does not represent an entire formation, especially when carbonate formations are present. Carbonate formations are characterized by their variation in porosity systems. Such porosity can be defined as intercrystalline, intergrain, moldic, vuggy and fracture and this parameter is linked with permeability values in different ways. In this work we compare the results obtained from the applications of neural network and fuzzy logic using conventional well logging, nuclear magnetic resonance data and information from core samples for permeability prediction. After using these two techniques, which can be considered efficient tools for uncertainty evaluation, a statistical coefficient called R2 shows better results when using logical fuzzy.