Model-Reduced Gradient Based Production Optimization

Abstract

Reduced-order modelling procedures, seen as an alternative to the classical gradient based approach, in which the adjoint of the tangent linear approximation of the forward model is replaced by the adjoint of a linear approximation of a reduced forward model. Proper orthogonal decomposition is used to compute the reduced model, by using the reduced adjoint, the gradient of the objective function can be approximated and the maximization problem can be solved in a reduced space. This work researches the viability of performing production optimization based on model-order reduction, by comparing several aspects of the implementation with similar aspects of Ensemble Optimization, a already well known way of solving the production optimization problem, that computes the gradient information from an ensemble of model realizations.