Modeling viscosity of light and intermediate dead oil systems using advanced computational frameworks and artificial neural networks

Abstract

The dead oil viscosity is a key parameter to numerous reservoir engineering problems such as modeling of (viscously-unstable) flow and transport in hydrocarbon reservoirs, sweep efficiency of enhanced oil recovery scenarios as well as the breakthrough times of the injected fluid. Prediction of this thermos-physical parameter, however, is of challenge due to nonlinear dependence on reservoir conditions as well as the crude oil characteristics. Previous studies have attempted to develop predictive empirical correlations or other intelligent models for dead oil viscosity; however, they often suffer from the lack of generality and required accuracy. In this work, based on a comprehensive databank from diverse geological sources, we develop three intelligent models –upon various schemes including simulated annealing programming, artificial neural network, and decision tree– for estimating dead oil viscosity. The latter may be used further for prediction of saturated and under-saturated oil viscosity as well. Our models function in wide range of temperatures and oil API gravity; hence, they can be employed as unified, general-in-purpose frameworks for universal prediction of dead oil viscosity. We compare the resulting novel frameworks with the pre-existing models available in the literature, and demonstrate the superiority of the decision tree-based model over others in terms of statistical (and graphical) error estimates as well as the (physical) validity of the model. The findings of this study can help for better understanding and more accurate management, simulation and prediction in different oil fields.