A hybrid assimilation scheme for characterization of three-phase flow in porous media. Issue 9 (2nd September 2019)

• Record Type: Journal Article
• Authors: Jahanbakhshi, Saman; Pishvaie, Mahmoud Reza; Bozorgmehry Boozarjomehry, Ramin
• Is Part Of: Inverse problems in science and engineering. Volume 27:Issue 9(2019)
• Page Start: 1195
• Page End: 1220
• Publication Date: 2019-09-02
• Subjects: Relative permeability -- absolute permeability -- joint estimation -- ensemble Kalman filter -- genetic algorithm; 76S05 -- 65M32 -- 62L12; Engineering mathematics -- Periodicals; Inverse problems (Differential equations) -- Periodicals; 620.001515357
• DOI: 10.1080/17415977.2017.1378196
• Languages: English
• ISSNs: 1741-5977

Abstract

ABSTRACT: In this study, ensemble Kalman filter (EnKF) is first applied to estimate absolute and relative permeabilities jointly under three-phase flow condition in the porous media. By assimilating historical data, absolute permeability field is adjusted progressively towards its reference. However, assimilation process does not improve the estimation of all relative permeability parameters, and some of them are poorly estimated at the end of assimilation. To improve the estimation of the relative permeability curves, we propose a new hybrid approach in which the estimation process of the absolute and relative permeabilities is separated. In this approach, gridblock permeabilities are again estimated within the EnKF framework in the inner loop. While, at each assimilation cycle, relative permeability parameters are updated in the outer loop using genetic algorithm so as to minimize the new proposed objective function. The proposed technique is validated in a synthetic two-dimensional reservoir model under three-phase flow condition. The hybrid approach results in monotonic reduction of the root-mean-squared error between the updated and the true permeability fields as well as accurate estimation of the relative permeability curves. Furthermore, the hybrid approach is compared against EnKF in terms of accuracy of the estimated model parameters, quantification of uncertainty, computational cost and quality of history-match.