Comparison between Multiple Linear Regression and Artificial Neural Networks for Porosity and Permeability Estimation

Porosity and permeability are two important characteristics of a hydrocarbon reservoir. The core measurements are usually used for these two parameter determination. This method is not only very expensive but also coring in many wells is not performable. Another method for this target is using of empirical equations. Any of these methods are associated with many problems. In addition, statistical methods will have some problems due to input data obtained from well logs. Artificial neural network is a new method, recently used in oil industry for prediction of petrophysical properties. This study is performed on Asmari Formation in Gachsaran oil field located in south of Iran. For porosity estimation two nets are used. One of these nets has 3 input parameters (density, sonic and neutron logs) and another has 10 input parameters (neutron, gamma, density, sonic, ILD, ILS, water saturation and spatial coordinate). Correlation coefficient between these nets predicted porosities and core porosities for generalization were 0.914 and 0.938 respectively and from multiple linear regression equation a 0.658 correlation coefficient is obtained. For permeability prediction two networks; one have six input parameters (density, sonic, gamma, ILD, ILS and porosity that obtained from porosity net with ten input parameters) and another with eleven input parameters (neutron, gamma, density, sonic, ILD, ILS, water saturation, spatial coordinate and porosity from porosity net with ten input parameters) and multiple linear regression equation are 0.851, 0.858 and 0.617 correlation coefficients are obtained respectively.