Fuzzy ARTMAP: a new tool for lithofacies recognition

In petroleum geology, lithofacies information is important for estimating porosity and permeability values from wireline logs at the un-cored well intervals; however, predicting lithofacies from logs is notan easy task. The results of lithofacies prediction from wireline log signals were compared using two different supervised classificationmethods: backpropagation neural network (BPNN) and a simplified version of Fuzzy ARTMAP called SFAM. The SFAM method gives results similar to those of BPNNs, but does not suffer BPNN's problems of excessivetraining time and the need for prior specification of network topology. If training time and the effort required to fine-tune BPNN parameters are acceptable, in some cases BPNN can provide significantly improved performance. However, to achieve this performance will generally require some degree of skill and a process of trial and error. Porosity and permeability predictions can be estimated by BPNNs; however,SFAM is currently unable to perform this task, and this requires further study. [ABSTRACT FROM AUTHOR]