Cluster Analysis in Reservoir Characterization




Muneta, Yasuhiro

Journal Title

Journal ISSN

Volume Title



Any raw data sampled in an oil field has a certain amount of noise; the sample may be called an obscure image of the real thing. We may eliminate the noise by a process of "image enhancement" in "statistical pattern recognition." Image enhancement is one of the important steps in processing large data sets to make them more suitable for classification than were the original data. In this work, cluster analysis, which is a method of image enhancement, is applied to some reservoir characterization problems such as permeability distributions of core samples, sand/shale sequences observed in wells, and pressure distributions in heterogeneous porous media to classify the sample data and find the intrinsic patterns (averaged images) from the original data sets. Cluster analysis is a multivariate statistical method. It is very general and can be applied to a wide area of scientific investigations. It is often called a tool of discovery or an unsupervised approach which doesn't depend on a priori information. It searches unknown-significant categories (patterns) themselves. Once we obtain typical patterns, we may analogously approach the real thing based on them. We find that cluster analysis is applicable to finding appropriate parent populations of a permeability distribution, theoretical indicator variograms of sand/shale sequences, and trends of effective permeability distribution.


LCSH Subject Headings