OLYMPUS optimization under geological uncertainty

摘要

Field development strategies are crucial for reservoir management, and over the last decade there has been quite some development of new optimization algorithms for solving this problem when the uncertainty in the reservoir description is provided by a set of reservoir models. To compare different approaches for this problem, the OLYMPUS benchmark challenge (Fonseca et al. 2018; TNO 2017) was defined, with three different tasks: well control optimization (task 1), field development optimization (task 2), and joint field development and well control optimization (task 3). This work presents solutions to the three exercises with two main optimization methods and problem-specific workflows. The main algorithms used in all three exercises are the ensemble-based optimization (EnOpt) and the line search derivative-free (LSDF) method. EnOpt is constructed for solving optimization problems where the uncertainty is represented by an ensemble of models, and in general it produced good results. However, we also found that the LSDF played an important role in quality checking the results obtained by EnOpt, and in some cases it provided superior results.