Derivative-free optimization for oil field operations

Abstract

A variety of optimization problems associated with oil production involve cost functions and constraints that require calls to a subsurface flow simulator. In many situations gradient information cannot be obtained efficiently, or a global search is required. This motivates the use of derivative-free (non-invasive, blackbox) optimization methods. This chapter describes the use of several derivative-free techniques, including generalized pattern search, Hooke-Jeeves direct search, a genetic algorithm, and particle swarm optimization, for three key problems that arise in oil field management. These problems are the optimization of settings (pressure or flow rate) in existing wells, optimization of the locations of new wells, and data assimilation or history matching. The performance of the derivative-free algorithms is shown to be quite acceptable, especially when they are implemented within a distributed computing environment. © 2011 Springer-Verlag Berlin Heidelberg.