Equation-of-state (EOS) fluid characterization is used to model the behavior of hydrocarbon reservoirs when variation in the fluid composition has a significant influence on the recovery of hydrocarbons. Examples are miscible gas-injection processes, where mass transfer between the injected gas and in-situ hydrocarbons can result in the injected fluid developing into a fluid that is miscible with the in-situ hydrocarbons, or gas-condensate fluids, where the liquid yield varies with pressure and composition as the reservoir depletes. For many reservoirs, more than one recovery mechanism is employed. For example, insufficient supply of miscible injectant might result in part of a reservoir employing miscible injection, while part is under waterflood. In such cases, accurate prediction of recovery for the miscible injection area might require an EOS characterization with a large number of components, while the phase behavior in the waterflood area could be modeled with sufficient accuracy with many fewer components. Phase-behavior calculations become much more computationally expensive as the number of components increases, but current commercial reservoir simulators must use the same number of components everywhere in the reservoir model. Thus, the requirement to use a large number of components to model the recovery process in part of the reservoir results in a large computational penalty in regions of the reservoir where a smaller number of components would suffice.
In this paper, we describe a method whereby the fluid can be locally lumped for phase-behavior calculations so that regions that are able to maintain sufficient compositional accuracy with fewer components can use less-expensive EOS calculations. Different lumpings can be used at different times in the life of the reservoir, as compositional effects become more or less important in different regions in the reservoir. Example problems that demonstrate the flexibility, efficiency, and consistency of the method are explored.