Compositional flow simulation is necessary for planning EOR (Enhanced Oil Recovery) and CO2 sequestration projects. We refer to the standard approach for performing the Phase Equilibrium (PE) computations as the Conventional-Variables (CV) method. In compositional simulation, the cost of PE calculations increases dramatically as the number of components increases, and the PE computations can become the most time consuming part of a simulation run. To reduce the number of components, lumping procedures are used, but that often entails the loss of possibly important phase-behavior information. The Reduced-Variables (RV) method is a robust and efficient scheme for PE calculations. However, there is limited information regarding the performance of RV in practical compositional flow simulation (i.e., coupled flow, transport, and thermodynamics). For example, there is only one published work with small 2D IMPEC (IMplicit Pressure, Explicit Compositions) simulations. Here, we integrated the RV method with a General Purpose Research Simulator (GPRS). This modular, object-oriented, computational framework allows us to make direct performance comparisons of RV- and CV-based compositional simulations. We use five different fluids, ranging from 6 to 26 components, and we focus on challenging multi-contact miscible displacements. We use two reservoir models: the first is the top layer of the SPE10 problem, and the second is a Discrete Fracture Model (DFM) with unstructured grids. The RV-based simulations are significantly more efficient than state-of-the-art CV-based computations. For the problems we describe here, the RV-based PE calculations can be up to seven times more efficient than the standard (CV-based) approach, and the total simulation time using RV can be a quarter of that used by the standard method. We show that the cost of the RV method for PE calculations is a linear function of the number of reduced variables, and we show that the total simulation time using RV is a weak function of the number of components used to describe the fluid. These properties of the RV approach enable us to study complex EOR processes in large-scale reservoir models, where large numbers of components and grid blocks are required. We propose that generalpurpose simulators use the RV approach for the PE computations associated with large-scale compositional simulations.

You can access this article if you purchase or spend a download.