Evaluation of practical scenario reduction approaches to robust derivative-free optimization for field development applications

Field development planning entails complex decisions to account for subsurface resources and surface infrastructure. Many of these decisions are of a discrete nature, typically categorical or combinatorial, and require solving mixed-integer optimization problems using dedicated numerical optimization approaches. However, these usually demand a large number of model evaluations to converge to an optimal solution. Model evaluations tend to be expensive and field development strategies are generally evaluated for an ensemble of model realizations to capture geological uncertainties, requiring multiple time-consuming reservoir flow simulations. As a result, the computational resources needed for optimization by standard discrete optimization approaches are prohibitively high. In this work we evaluate practical approaches to employ derivative-free optimization under uncertainty (robust optimization). We consider a case based on a synthetic oil field, where the drilling order of 6 wells is to be determined to maximize project economics. The underlying geological uncertainties are represented by an ensemble of 100 model realizations. To alleviate the computational load, we use adaptive scenario reduction approaches to find representative subsets of the full ensemble that will preserve the quality of the optimization results. Several approaches based on ranking performance, sampling probability and quantile selection are evaluated to select static and adaptive subsets during the optimization process for different derivative-free optimization methods (e.g., mesh adaptive search, genetic algorithm). Their performance is compared against results of optimization over the full ensemble, and validated using an exhaustive global search. The results show that a sizable reduction in computational cost can be achieved in terms of the number of required reservoir simulations using representative subsets without significantly compromising the optimization performance. The findings of this study also provide quantitative insights to help identify robust and effective scenario reduction approaches for realistic simulation-based field development optimization studies. This enables derivative-free optimization as a practical tool for field development planning, unlocking additional value from the use of subsurface resources. Copyright© 2024 by the European Association of Geoscientists & Engineers (EAGE). All rights reserved