Efficient Uncertainty Quantification and History Matching of Large-Scale Fields Through Model Reduction

Abstract

Uncertainty quantification (UQ) and history matching (HM) have become a regular routine for reservoir management and decision-making in petroleum industry. The zonation method was widely used to reparametrize correlated fields with one lumped constant or multiplier specified for each zone such that the dimensionality of problems can be reduced and the HM problem can be efficiently solved. However, this ad hoc method faces a challenge to find the optimal zones. Moreover, it may fail to honor the geological (or geostatistical) features after the lumped constants or multipliers are applied, resulting in patches. In this work, we present several PCA-based techniques to address this problem by reducing the dimensionality of problem but not subject to the limitations of the zonation method.