Using an Artificial Neural Network for Predicting Water Saturation in an Iranian Oil Reservoir

Abstract

Water saturation is defined as that fraction, or percentage, of the pore volume occupied by water. Estimation of the water saturation greatly impacts oil-in-place calculations. A saturation height function is used to predict the saturation in the reservoir for a given height above the free water level. Traditionally, methods for predicting water saturation as a function of rock properties and height above contact have fallen into two groups, those based on capillary pressure curve averaging and log-based methods. In this study, eight saturation height methods (i.e., Leverret [1941], cap-log, Cuddy et al. [1993], Johnson [1987], modified cap-log [2007], modified Cuddy [2007], Skelt-Harrison [1995], and Sodena methods) employed in the oil industry are investigated. In this article, a new artificial neural network (ANN) is developed to predict the water saturation. Two hundred sixty-three data sets were collected from a southern Iranian reservoir. Porosity (Φ), permeability (k), and height above the free water level (h) were used as the input data and water saturation was the target data. Seventy percent of these data points were used for training and the remainder for predicting the Sw (validation and test). An ANN was developed and a correlation coefficient (R2) of 0.985 and absolute average relative error (%) of 6.40% were obtained by comparing water saturation obtained from the ANN model and logging.