Water Saturation and Porosity Prediction Using Back-propagation Artificial Neural Network (BPANN) from Well Log Data

Abstract

Porosity and water saturation are two fundamental parameters in reservoir characterization. In this study, for predicting both mentioned parameters artificial neural network was used as intelligent technique.  Five variables include neutron log, effective porosity, caliper log, bulk density, and sonic log were used from 3 wells from one of the Iranian oil fields. After normalizing data Seventy percent of data were used as training dataset and remainder for testing the network. Several feed –forward neural networks were operated to obtain best performance of different algorithms to train the network. Levenberge-Marquardt back-propagation algorithm was chosen as the training algorithm which had the best performance and was faster than other algorithms. Optimum neurons in the hidden layer for porosity and water saturation were obtained respectively. Results shown that Backpropagation artificial neural network (BPANN) has a high ability to predict porosity and water saturation which correlation between real output and predicted output using BPANN were 0.82 and 0.93 respectively.