Investigating the effect of correlation-based feature selection on the performance of neural network in reservoir characterization

Abstract

Accurate prediction of permeability is very important in characterization of hydrocarbon reservoir and successful oil and gas exploration. In this work, generalization performance and predictive capability of artificial neural network (ANN) in prediction of permeability from petrophysical well logs have been improved by a correlation-based feature extraction technique. This technique is unique in that it improves the performance of ANN by employing fewer datasets thereby saving valuable processing time and computing resources. The effect of this technique is investigated using datasets obtained from five distinct wells in a Middle Eastern oil and gas field. It is found that the proposed extraction technique systematically reduces the required features to about half of the original size by selecting the best combination of well logs leading to performance improvement in virtually all the wells considered. The systematic approach to feature selection eliminates trial and error method and significantly reduces the time needed for model development. The result obtained is very encouraging and suggest a way to improve hydrocarbons exploration at reduced cost of production. Furthermore, performance of ANN and other computational intelligence techniques can be improved through this technique.