Abstract
Reservoir simulation workflows contain significant elements of uncertainty, particularly in the geological
description of reservoir geometry and petrophysical parameters such as permeability and porosity. To
accurately account for uncertainty and span the range of likely outcomes, different equiprobable realiza-
tions should be kept as long as possible throughout a modelling workflow. However, working with
multiple realizations of the same reservoir for optimization purposes may quickly become prohibitively
expensive, particularly since using a full forward simulation can be quite demanding even for a single
model realization. Herein, we propose to combine two recent and quite different technologies to enable
optimization of multiple realizations. The first is the use of multiscale technology, wherein approximate,
but well-behaved pressure solutions can be efficiently computed using precomputed basis functions that
capture local flow features. Secondly, the use of single-phase flow diagnostics can serve as an efficient
alternative to full physics flow simulations for optimization and characterization purposes. By combining
these technologies in a single implementation, we obtain a workflow that makes it possible to quickly
evaluate and optimize multiple realizations, while still retaining error control. In particular, it is possible
to adjust accuracy dynamically from inexpensive proxy models provided by pure multiscale and flow
diagnostics, via more accurate iterated multiscale solutions and incompressible flow, to fully-implicit
solvers that incorporate the relevant flow physics.
description of reservoir geometry and petrophysical parameters such as permeability and porosity. To
accurately account for uncertainty and span the range of likely outcomes, different equiprobable realiza-
tions should be kept as long as possible throughout a modelling workflow. However, working with
multiple realizations of the same reservoir for optimization purposes may quickly become prohibitively
expensive, particularly since using a full forward simulation can be quite demanding even for a single
model realization. Herein, we propose to combine two recent and quite different technologies to enable
optimization of multiple realizations. The first is the use of multiscale technology, wherein approximate,
but well-behaved pressure solutions can be efficiently computed using precomputed basis functions that
capture local flow features. Secondly, the use of single-phase flow diagnostics can serve as an efficient
alternative to full physics flow simulations for optimization and characterization purposes. By combining
these technologies in a single implementation, we obtain a workflow that makes it possible to quickly
evaluate and optimize multiple realizations, while still retaining error control. In particular, it is possible
to adjust accuracy dynamically from inexpensive proxy models provided by pure multiscale and flow
diagnostics, via more accurate iterated multiscale solutions and incompressible flow, to fully-implicit
solvers that incorporate the relevant flow physics.
