Author(s): |
J. J. Gómez-Hernández, Universitat Politècnica de València (ES)
H. Zhou, The University of Texas at Austin (US)
L. Li, The University of Texas at Austin (US)
H.-J. Hendricks Franssen, Forschungszentrum Jülich GmbH (DE)
|
Abstract: |
Recently, in the fields of hydrogeology and petroleum engineering, a big effort has been made in the development of techniques for the characterization of geological formations from sparse data. The aim of the latest developments is the generation of realizations of the geological architecture that display realistic patterns of variability, such as those observed in outcrops. This characterization is only possible using non-multiGaussian statistical models. Also recently, in these same fields, a big effort has been made in inverse modeling (or history matching) with emphasis in the generation of multiple realizations of the parameters that control flow and transport in the sub-surface. All of these realizations must be conditional on the measurements of the state variables, in the sense, that the numerical solution of the state equation in these realizations predicts correctly the state at measurement locations. But, there is no efficient technique capable of generating non-multiGaussian realizations that are, at the same time, conditional to state-variable measurements. The objective of this work is to present such a technique through a fusion of multiple point geostatistics and ensemble Kalman filtering, since they are the most efficient techniques for non-Gaussian characterization and stochastic inverse modeling, respectively |
