Derivative-free trust region optimization for robust well control under geological uncertainty

Abstract A Derivative-Free Trust-Region (DFTR) algorithm is proposed to solve the robust well control optimization problem under geological uncertainty. (DF) methods are often a practical alternative when gradients not available or unreliable due cost function discontinuities, e.g., caused by enforcement of simulation-based constraints. However, effectiveness DF for solving realistic cases heavily dependent on an efficient sampling strategy since calculations involve time-consuming reservoir simulations. The DFTR samples space around incumbent solution and builds quadratic polynomial model, valid within bounded region (the trust-region). minimization model guides method in its search descent. Because curvature information provided model-based routine, trust-region approach able conduct more compared other methods, direct-search approaches. implemented FieldOpt, open-source framework field development optimization, tested Olympus benchmark against two types commonly applied production optimization: (Asynchronous Parallel Pattern Search) population-based (Particle Swarm Optimization). Current results show that has improved performance model-free In particular, presented convergence, being capable reach solutions with higher NPV requiring comparatively fewer iterations. This feature can be particularly attractive practitioners who seek ways improve strategies while using ensemble full-fledged models, where good convergence properties even relevant.