Publications

This paper describes the use of backpropagation neural networks (BPNNs) to predict permeability profiles from well logs in four wells located in Sumatra, Indonesia. The reservoirs are located in the Telisa Formation with significant heterogeneity. We develop a BPNN permeability estimator for each of the cored wells and validate the results in other wells. The approach does not only give better estimates than the Wyllie & Rose equation, it also provides maximum and minimum permeability estimates. We use the range value (maximum minus minimum values) as a reliability indicator, which is particularly useful for examining the effects of reservoir heterogeneity and data sampling on permeability estimates.