Facies Classification in Basaltic Volcanic Rocks Using Artificial Neural Networks
Indrajit Bandyopadhyay, Sangeeta Singhal Bandyopadhyay, and Dedi Juandi
DCS, Schlumberger, Beijing, China
Basaltic volcanic rocks generally are homogeneous in composition and devoid of crystal size variation in specific geologic setting. Compositional and textural homogeneity hinders classification of reservoir quality in basaltic rocks. Mode of genesis and tectonics impart morphological signatures, which differentiate basaltic rocks into reservoirs with varying hydrocarbon potential. Units with fractures and vesicles are better reservoirs than those which are glassy, massive or heavily brecciated. High-resolution azimuthal borehole images allow identification of morphological features. Image logs can be used to classify basalts into morphofacies based on relative abundance of these features. Morphofacies classification including Pillow Lava, Intrusive Lava, Sheet Lava, Massive Lava, Bedded Lava and Brecciated Lava, identified from images in a key well can be extended to other wells using artificial neural network (ANN). ANN is designed using a facies model and set of discriminator logs. Facies model is created with morphofacies identified from key well. Discriminator logs are selected so that the responses correlate with morphofacies variation and include spectral GR, resistivity, bulk density, neutron porosity, photoelectric factor and sonic transit time. The ANN is then trained over selected intervals of key well, with remaining intervals being used to validate its performance. Once trained, it can be used to define facies in wells without image logs but having discriminator logs. Vertical facies assemblages within a basaltic sequence from multiple wells, will portray lateral facies distribution and therefore reservoir property variation. The paper illustrates the process and results of performing morphofacies classification in basaltic rocks from images in two wells followed by propagation of this classification in four wells having only discriminator logs using ANN.
Presentation GEO India Expo XXI, Noida, New Delhi, India 2008©AAPG Search and Discovery