Home

Quick Links

Search

 
A Comparative Study of Reduced Variables Based Flash and Conventional Flash (SPE 154477)Normal access

Authors: E. Stenby and W. Yan
Event name: 74th EAGE Conference and Exhibition incorporating EUROPEC 2012
Session: Reservoir Simulation II - Numerical (Europec)
Publication date: 04 June 2012
Organisations: SPE, EAGE
Language: English
Info: Extended abstract, PDF ( 385.69Kb )
Price: € 20

Summary:
Speeding up flash calculation is a central issue in compositional reservoir simulations since phase equilibrium calculation is the most time-consuming part in those simulations. The reduced variables methods, or the reduction methods, reformulate the original phase equilibrium problem with a smaller set of independent variables. Various versions of the reduced variables methods have been proposed since the mid 80's. The methods were first proposed for cubic equations of state (EoS) with zero binary interaction parameters (BIPs) and later generalized to situations with non-zero BIP matrices. Most of the studies in the last decade suggest that the reduced variables methods are much more efficient than the conventional flash method. However, Haugen and Beckner questioned the advantages of the reduced variables methods in their recent paper (SPE 141399). A fair comparison between the reduced variables based flash and the conventional flash is not straightforward since the former is difficult to be formulated as unconstrained minimization and involves more complicated composition derivatives. With the recent formulations by Nichita and Garcia (2010), it is possible to code the reduced variables methods without extensive modifications of MichelsenC"s conventional flash algorithm. A minimization based reduced variables algorithm was coded and compared with the conventional minimization based flash. A test using the SPE 3 example showed that the best reduction in time was less than 20% for the extreme situation of 25 components and just one row/column with non-zero BIPs. A better performance can actually be achieved by a simpler implementation directly using the sparsity of the BIP matrix.


Back to the article list