TY - JOUR

T1 - Speeding up the flash calculations in two-phase compositional flow simulations - The application of sparse grids

AU - Wu, Yuanqing

AU - Kowitz, Christoph

AU - Sun, Shuyu

AU - Salama, Amgad

N1 - KAUST Repository Item: Exported on 2020-10-01

PY - 2015/3

Y1 - 2015/3

N2 - Flash calculations have become a performance bottleneck in the simulation of compositional flow in subsurface reservoirs. We apply a sparse grid surrogate model to substitute the flash calculation and thus try to remove the bottleneck from the reservoir simulation. So instead of doing a flash calculation in each time step of the simulation, we just generate a sparse grid approximation of all possible results of the flash calculation before the reservoir simulation. Then we evaluate the constructed surrogate model to approximate the values of the flash calculation results from this surrogate during the simulations. The execution of the true flash calculation has been shifted from the online phase during the simulation to the offline phase before the simulation. Sparse grids are known to require only few unknowns in order to obtain good approximation qualities. In conjunction with local adaptivity, sparse grids ensure that the accuracy of the surrogate is acceptable while keeping the memory usage small by only storing a minimal amount of values for the surrogate. The accuracy of the sparse grid surrogate during the reservoir simulation is compared to the accuracy of using a surrogate based on regular Cartesian grids and the original flash calculation. The surrogate model improves the speed of the flash calculations and the simulation of the whole reservoir. In an experiment, it is shown that the speed of the online flash calculations is increased by about 2000 times and as a result the speed of the reservoir simulations has been enhanced by 21 times in the best conditions.

AB - Flash calculations have become a performance bottleneck in the simulation of compositional flow in subsurface reservoirs. We apply a sparse grid surrogate model to substitute the flash calculation and thus try to remove the bottleneck from the reservoir simulation. So instead of doing a flash calculation in each time step of the simulation, we just generate a sparse grid approximation of all possible results of the flash calculation before the reservoir simulation. Then we evaluate the constructed surrogate model to approximate the values of the flash calculation results from this surrogate during the simulations. The execution of the true flash calculation has been shifted from the online phase during the simulation to the offline phase before the simulation. Sparse grids are known to require only few unknowns in order to obtain good approximation qualities. In conjunction with local adaptivity, sparse grids ensure that the accuracy of the surrogate is acceptable while keeping the memory usage small by only storing a minimal amount of values for the surrogate. The accuracy of the sparse grid surrogate during the reservoir simulation is compared to the accuracy of using a surrogate based on regular Cartesian grids and the original flash calculation. The surrogate model improves the speed of the flash calculations and the simulation of the whole reservoir. In an experiment, it is shown that the speed of the online flash calculations is increased by about 2000 times and as a result the speed of the reservoir simulations has been enhanced by 21 times in the best conditions.

UR - http://hdl.handle.net/10754/564087

UR - https://linkinghub.elsevier.com/retrieve/pii/S0021999115000169

UR - http://www.scopus.com/inward/record.url?scp=84921637066&partnerID=8YFLogxK

U2 - 10.1016/j.jcp.2015.01.012

DO - 10.1016/j.jcp.2015.01.012

M3 - Article

SN - 0021-9991

VL - 285

SP - 88

EP - 99

JO - Journal of Computational Physics

JF - Journal of Computational Physics

ER -