It is normally understood in the industry that permeability and porosity are related but relationship can be strong or weak based on heterogeneity, anisotropy and lithology of the reservoir. So far linear models have been arbitrarily being tried to fit and matched with the data thereby resulting in a poor relationship. This paper seeks to explore data mining techniques for pattern recognition including Kohonen self-organizing maps and other unsupervised learning techniques.
Kohonen self-organizing maps and principal component analysis are applied over permeability –porosity data to first cluster the data into various sets. This data can then be linked back to the formations and a predictor can be built thereby giving a tool for reservoir characterization. Then a linear model is to be built with the linearity being assessed by Shapiro Wilk Test and Quantile – Quantile (Q-Q) plot.
This technique develops a relationship with a much better fit between permeability and porosity in respective ranges and thereby creating a higher cue-efficient of determination i.e. R squared value.