Quick Search:    advanced search

GSW Home   GeoRef Home   My GSW Alerts   Contact GSW   About GSW   Journals List   Help

Petroleum Geoscience

This Article Figures Only Full Text Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Bhatt, A. Articles by Helle, H. B. Search for Related Content GeoRef GeoRef Citation

Determination of facies from well logs using modular neural networks

¹ Norsk Hydro ASA, E&P Research Centre, N-5020 Bergen, Norway (e-mail:Hans.B.Helle{at}nho.hydro.com)
² Norwegian University of Science and Technology, Department of Petroleum Engineering and Applied Geophysics, N-7491 Trondheim, Norway (Present address: Statoil ASA, PO Box 273, N-7501 Stjørdal, Norway (e-mail:ALBH{at}statoil.com))

Zonation of well logs and the correlation of zones between wells are primary tasks in sub-surface geological and engineering analysis. We propose in this paper an artificial neural network (ANN) approach for objective clustering and identification of facies from well logs. The method relies upon combining back-propagation neural networks in ensembles and modular systems, where the multi-class classification problem of facies identification has effectively been reduced to a number of two-class problems.

Based on the neural network responses using synthetic logs from a realistic model, we optimized the architecture and training procedure of the component networks in the modular system, where the building blocks are simple three-layer back-propagation ANNs. Ensembles of ANNs are trained on disjoint sets of patterns using soft overtraining to ensure diversity and generalization. Recurrent ANNs are shown to enhance the facies continuity by effectively removing ambiguous or spurious classifications.

The performance of the technique was demonstrated using synthetic data and it was then used to detect four different facies within the Ness Formation in the North Sea. An average hit rate of above 90% in wells not used for training the network is slightly to significantly better than the performance published for similar classification experiments.

KEYWORDS: lithology, neural network, well log interpretation