Chaotic optimization for reservoir history matching

Computer-aided history matching of oil production data requires the solution of a challenging minimization problem. Usually, gradient based techniques have been applied to this aim: starting from an initial guess of the solution, these methods converge to the closest local minimum, along specified descend directions. Typically, however, the parameter space is populated by many local minima, each giving a different response in terms of forecasted production profiles. Theoretically, it is thus important to identify the lowest possible minimum among all candidates. In this paper we propose the application of the chaotic optimization technique to reservoir history matching problems. We show that this approach can highly speed up the solution process by providing optimal starting points for a convex minimization routine to be applied in sequel.