Reservoir Characterization Using Distance Based Ensemble Smoother with Permeability Distribution Pattern

Abstract

A distance is the degree of model dissimilarity and it is important for effective model selection. This paper suggests a cross spatial pattern to find permeability distribution from an injector to a producer. The distance is defined as one minus correlation coefficient of permeability data obtained by the spatial pattern. Using multi-dimensional scaling, initial 400 reservoir models are projected on two dimensions based on the distance. By K-medoids clustering, they are classified into 10 groups. One representative medoid is chosen with the least difference in productions from the reference field. Then, 100 models are selected around the medoid for ensemble smoother(ES). The proposed distance can achieve improved reservoir characterization and history matching combined with ES. Also, this method helps to reduce uncertainty ranges of future oil and water productions, and decreases total simulation time by 75% with proper sampling of good 100 models.