Estimating Permeability by Use of Neural Networks in Thinly Bedded Shaly Gas Sands

Abstract

Summary In this paper, we develop methods of estimating permeability that use conventional log and nuclear magnetic relaxation (NMR) log data to estimate both computed permeability (by NUMAR Inc.) and Klinkenberg core-permeability measurements. The permeability estimation methods we use in this work are based on multilayer neural networks. We compare the proposed methods of estimating permeability to traditional methods in a thinly bedded siliciclastic gas reservoir in the U.S. The results of this comparison demonstrate that neural networks can be used successfully to transform conventional or NMR logs into desired permeability. The three different permeabilities predicted by use of appropriate neural networks operate by transforming (1) conventional log data into NUMAR-derived permeability, (2) conventional log data into Klinkenberg permeability, and (3) NMR echoes into Klinkenberg permeability.