Abstract

In order to develop the flash calculation procedure for EOS-compositional reservoir simulators, the diagnostic phase, and combination of numerical methods  are reviewed in order to determine most efficient scheme in terms of run-time, memory and CPU usage, testing different sets of possible initial guesses, direct flash calculation vs. stability test and phase boundary problems.

Some significant conclusions are obtained: initial guess plays a very important role; a diagnostic phase is needed prior to starting the calculations; a stability analysis phase may reduce the run-time even if there is an initial guess; the best choice for stability initial guess is the converged flash k-values from a previous time-step; the initial guess is not usually good enough to start with the QN methods directly; the numerical method chosen should be determined by the diagnostic phase; an intelligent decision-making engine can greatly decrease the run-time by switching between SSA, NR and QN, which can also decrease the probability of unwanted solutions; an existing acceleration method for multiphase flash is corrected.