Development of a multi-formulation compositional simulator

Compositional simulation is a complex task that involves solving several equations simultaneously for all grid blocks representing a petroleum reservoir. Usually, these equations are separated into two groups: primary and secondary equations. Similarly, the unknowns of the system are also separated into primary and secondary variables. Considering the large number of unknowns, there are many ways to separate such variables in order to deal with the primary variables. This work aims at comparing a number of formulations for compositional reservoir simulation. It also aims at enhancing the formulations with new features not provided in the original publications. To accomplish these objectives, various formulations prevailing in the literature are implemented in The University of Texas at Austin in-house fully implicit simulator named GPAS (General Purpose Adaptive Simulator) and their performances were compared. Subsequently, some of the formulations were enhanced and tested for various applications. The comparison of the formulations studied indicated differences in efficiency for each approach. These differences come from the fact that when one is solving for a different set of primary variables, the manipulation of the equations is analogous to the use of a preconditioner applied to a linear system of equations. Furthermore, unlike a preconditioner, changing the primary variables affects the non-linear solver. Therefore, differences in terms of the number of Newton-Raphson iterations, used for solution of nonlinear equations resulting from discretization of nonlinear partial differential equations representing fluid flow in the reservoir, are expected. In addition to these differences in the non-linear solver, many formulations explore the fact that a reduced number of equations need to be solved implicitly, thus considerably reducing the CPU time dedicated to the linear solver. Finally, new features not provided in the original published formulations such as three-phase flash calculation, physical dispersion, and unstructured grid were implemented and verified. Additionally, it was demonstrated that, in certain situations, these enhancements are essential to properly model the physical phenomena occurring in oil and gas reservoirs.