Finite analytic numerical method for three-dimensional fluid flow in heterogeneous porous media

Abstract

Understanding fluid flows in heterogeneous porous media is fundamental to applied geosciences. The wide connectivity variations in the natural aquifer or oil reservoirs make the equivalent permeability have strong spatial variations. When performing the simulations for subsurface flows, the permeabilities may have strong discontinuities across the interfaces between different grid cells. Utilizing the traditional numerical schemes to simulate flows in strong heterogeneous media, the refinement ratio for the grid cell needs to increase dramatically to get an accurate result. Recently, we proposed a finite analytic numerical scheme to solve the two-dimensional fluid flows in heterogeneous porous media. With only 2×2 or 3×3 subdivisions, this scheme can provide rather accurate solutions. In this paper, we develop the finite analytic numerical method for solving the three-dimensional fluid flows in heterogeneous porous media. For the rectangular grid system, it is generally proposed that the pressure gradient in a plane normal to the edge joining different permeability regions will tend to infinite as approaching the edge according to a typical power-law solution and the tangential derivate of the pressure along the edge must be of limited value due to the pressure continuity. Consequently, the three-dimensional flow will reduce to the two-dimensional one in the neighborhood around each edge. Such quasi-two-dimensional behavior is then applied to construct a finite analytic numerical scheme. Numerical examples show that the proposed scheme can provide rather accurate solutions with only 2×2×2 or 3×3×3 subdivisions and the convergent speed is independent of the permeability heterogeneity. Due to its high calculation efficiency, the proposed scheme is utilized to test the well known LLM (Landau, Lifshitz and Matheron) conjecture, which provides keq/kG=exp⁡(16σln⁡k2) for the isotropic log-normal porous medium. The numerical results do not support this conjecture for large σln⁡k, but strongly suggest the linear relation keq/kG=1+16σln⁡k2.