Uses and abuses in applying neural networks for predictions in hydrocarbon resource evaluation

Abstract

Abstract Artificial neural network (ANN) is a powerful tool for reservoir characterization; it can be used for classifications of categorical reservoir variables, as well as for predictions of continuous reservoir properties. Yet it also has a number of drawbacks, including the deficiency for physical interpretations of results, and difficulties in tuning the training and iteration parameters. This article presents uses and limitations of ANN for lithofacies clustering from wireline logs and for porosity prediction in 3D reservoir modeling from seismic attributes and geological interpretations. Through a number of examples, it demonstrates the importance of integrating ANN with the physics-based model and insightful knowledge of subject problems. The optimal methodology generally is to combine ANN with statistical and/or geostatistical methods in predicting reservoir properties.