Features Modeling of Oil Sands Reservoirs in Metric Space

Abstract

Uncertainty in oil sands reservoirs can be quantified by generating multiple realizations using geostatistical methods. However, it requires huge computing time to simulate all of the realizations. This article proposes a new approach for features modeling of oil sands reservoirs in metric space. As the first step, an area affected by the expansion of a steam chamber is set and converted to the polar coordinate system. The converted area is expressed as an image matrix consisting of 0 or 1 value. Then the matrix is transformed using two-dimensional discrete Fourier transform. Key features in the front columns and rows of the transformed matrix are extracted. These features in metric space are plotted using principal component analysis. Self-organizing map algorithm is used to select representative models of realizations for performing full flow simulations. In the result of grouping, each cluster group distributes separately in metric space according to reservoir productivity, but there are mixes of a small portion among the adjacent groups due to similar productivity.