Simulação numérica da recuperação avançada de petróleo: abordagem em CFD e redes neurais artificiais

Abstract

The global energy demand is growing and to accompany it, the oil industry uses different recovery techniques. Among these, there are the so-called enhanced oil recovery methods, which are methods that aim to produce the oil contained in the reservoir, which cannot be produced in an economically viable way by conventional strategies of water and immiscible gas injection. In this sense, petroleum engineering has benefited from nanoscience, with contributions in the reservoir area, in which nanofluids, with their special chemical and physical characteristics, are capable of transforming the porous medium and displacing oil much more efficiently than the secondary recovery method. To accomplish these feats, it is necessary an approach that combines economic and environmental aspects, and for that, this work aimed to evaluate the efficiency of the CaCO3-based nanofluid, which originates from the reuse of sururu shells, in the displacement of oil through computational fluid dynamic analysis comparing the recovery factor obtained by the water injection recovery method with the recovery factor obtained by the enhanced nanofluid injection method. Furthermore, an artificial neural network with feed-forward architecture was developed that predicts the recovery factor, based on data related to the injection method, characteristics of the porous media, and oil properties. The results showed an increase in the oil recovery factor with the increase in the concentration of nanoparticles, being superior to the recovery factor obtained by the injection of water. In addition, the neural network was successful in predicting the recovery factors from different characteristics of the porous medium, the oil and the injected fluid, with R² of 0.887 and a mean square error of 0.3362.