A realization-based approach for uncertainty quantification using distances and kernel methods

@inproceedings{ScheidtRM2007,
 abstract = { Assessing uncertainty in reservoir performance requires the analysis of a large number of parameters. To capture the wide range of uncertainty in flow response, a large set of realizations should be processed. Geostatistical algorithms can rapidly provide multiple, equally probable realizations. However, due to the large computational demand of flow simulations, only a small number of realizations can be simulated in practice. Static properties of the realizations are therefore often used for ranking realizations. Traditional ranking techniques for selection of P10, P50 and P90 flow responses are highly dependent on the static property used (e.g. OOIP). In this paper, we propose a new method to parameterize spatial uncertainty on a large set of reservoir models. Based on a single parameter, the distance between any two realizations, the method consists in mapping each realization in a Euclidean space using a multidimensional scaling technique. The distance function should be tailored to the particular problem, in this case, flow responses. Kernel methods, such as Kernel Principal Component Analysis or Kernel k-means are then used in the Euclidean space in order to select a few representative realizations. This small subset of models should have the same properties, in terms of flow, than the entire set. Full flow simulations can then be performed on the small set of realizations for uncertainty quantification (e.g. P10 P50 and P90 quantiles estimation). The efficiency of this new method is demonstrated in a synthetic case with channel facies. },
 author = { Scheidt, Celine AND Caers, Jef },
 booktitle = { 27th gOcad Meeting },
 month = { "june" },
 publisher = { ASGA },
 title = { A realization-based approach for uncertainty quantification using distances and kernel methods },
 year = { 2007 }
}