Abstract
Simulation of isothermal fluid flow in a reservoir using a compositional simulator requires fluid properties that are functions of pressure and properties that are functions of pressure and composition. These properties, i.e., K-values, densities and viscosities of both vapor and liquid phases, are usually obtained from general correlations phases, are usually obtained from general correlations or laboratory measurements of a reservoir fluid sample during a differential-depletion experiment in a PVT cell. prediction of fluid properties of complex mixtures using existing correlations is generally subject to great uncertainties. The laboratory measured data that are generally correlated as functions of pressure have validity only over a limited range of compositional variation. The purposes of this paper were (1) to assess, using a linear compositional simulator, the error introduced into calculated reservoir performance by employing fluids with a given range of uncertainties in their physical properties; and (2) to examine the validity of using the physical data correlated in the compositional simulator as functions of pressure rather than functions of both pressure and composition. The gas cycling process was chosen for illustration because composition changes during this process are large and results are affected more than in a depletion-type process. The hypothetical reservoir fluid system considered in this study was a methane-n-butane-n-decane mixture chosen to simulate a volatile oil system. The results of this investigation show for the particular system studied that:(1)the K-values for particular system studied that:(1)the K-values for the lighter components have the most significant effect on the calculated reservoir performance; and(2)simulations using fluid properties that are equivalent to the data measured during a differential depletion experiment reliably predict reservoir performance even under conditions where significant performance even under conditions where significant variations in reservoir fluid composition occur.
Introduction
A number of papers have recently been published concerning the development of compositional reservoir simulators-the mathematical models that simulate isothermal flow of multiphase, multicomponent fluids in porous media considering mass transfer effects. These models, which properly describe the distribution of each individual component in both vapor and liquid phases and account for pressure and compositional dependence of K-values, phase densities and viscosities, are more rigorous than the conventional simulators. The latter assumes that the heavy component does not exist in the vapor phase. To use the compositional simulator, it is highly desirable that fluid properties, i.e., K-values, densities and viscosities, as functions of pressure and composition, be available. However, for complex reservoir fluid mixtures, this information is rarely available. These fluid properties are usually calculated from published generalized correlations or obtained from laboratory measurements of a reservoir fluid sample by performing differential depletion experiments in a PVT cell. Prediction of fluid properties of complex mixtures using existing correlations is generally subject to great uncertainty. These errors will certainly have effects on the predicted reservoir performance. These effects may predicted reservoir performance. These effects may even be amplified if all the fluid properties are calculated from correlations. Improvement of the correlation predicted data by adjusting these data to match the limited available experimental values for the system of interest can be make. Yet there is no guarantee that the adjusted data will describe reliable fluid behavior in the region away from the matched points. On the other hand, the laboratory measured data, which are expressed as functions of pressure only, have validity over a limited range of pressure only, have validity over a limited range of compositional variation. When compositions of reservoir fluids vary significantly, the reliability of applying the laboratory measured data in the numerical simulation becomes questionable.
SPEJ
p. 3
