ABSTRACT

Data mining, also known as knowledge discovery in databases, is an information processing technology for extracting higher-level knowledge (such as rules, concepts, or regularities) from lower-level data (such as well logs or petrographic characteristics of rocks). A data mining approach was proven useful for the prediction of porosity and permeability for a given reservoir.

This approach employs an algorithm that can automatically construct the "if-then" rules that relate well logs to porosity (or permeability) for the reservoir of interest. First, a generator builds an entropy-based decision tree, which in turn can be pruned, split, or lumped before the rule extraction procedure. The results are in the form of if-then rules, thereby allowing the porosity and permeability to be predicted by a rule inference procedure. This approach has been praised for its high knowledge comprehensibility and transparency, which allows geoscientists to judge, modify, and interpret the characteristics of a given reservoir.

A field site, Appleton Field located in north central Escambia County, Alabama, was studied. Data from one well was used as the training set to extract the rules and data from four other wells was used to test the algorithm. The results of these predicted values were compared with the available core data. The average error rates of the porosity and permeability predictions for the test wells are 18% and 12%, respectively. The result generated by the proposed algorithm was compared with results using Stratamodel (Landmark Graphics Corp.) software. The results given by the proposed approach was found to have higher accuracy than those from Stratamodel.