Data integration for electrofacies classification using self-organizing maps

Abstract

In order to minimize uncertainties inherent to electrofacies classification from well log data, this paper describes nine different strategies for classification using unsupervised neural networks, self-organizing maps (SOM), which combine geophysical information with data derived from well logs, Hölder exponent and acoustic impedance. The method was applied to the Albian carbonate reservoir of the Campos Basin, for which the following well logs were used: sonic (DT), neutron porosity (NPHI), density (RHOB) and gamma ray (GR). Due to the scarcity of core data, the classifications were performed in two steps: first, the classification of well logs and acoustic impedance were performed, which were then used as an additional variable in tests with the algorithm seeking to classify the core data that escribe four core data facies (reservoir, possiblereservoir, non-reservoir and cement). The best results of the analysis are associated with the insertion of acoustic impedance information and of such new variable. Adding the new variable to well log samples in the training dataset resulted in a 16% increase in accuracy in core data classification. The results allow the potential integration of seismic data to be quantified in the automatic classification of well data by the SOM method.