Use of Linearized Reduced-order Modeling and Pattern Search Methods for Optimization of Oil Production

Computational optimization holds great promise for the management of oil field operations. These optimizations can be very expensive computationally, however, because each function evaluation requires a reservoir simulation, which is itself time consuming. In this paper, we present and apply a surrogate modeling procedure which greatly accelerates optimizations based on subsurface flow simulations. The surrogate model, a trajectory piecewise linearization (TPWL) technique, represents new states in terms of expansions around states simulated during training runs (the training runs require full-order simulations). The TPWL representation is incorporated into a generalized pattern search optimization procedure. Results for example problems demonstrate significant improvement in the objective function and a two order of magnitude reduction in the number of full-order simulations required for the optimization.