Vertical lithological proxy using statistical and artificial intelligence approach: a case study from Krishna-Godavari Basin, offshore India

We have identified the lithologies from well logs and available core data through the cluster and neural network analysis in the Krishna-Godavari (KG) basin. The unsupervised hierarchical cluster analysis (HCA) has been used to find out the dissimilarity behaviour of pairwise well log data associated with each lithological unit as a measure of proximity. Whereas, supervised neural network analysis has been used for the identification of lithologies of un-cored portion of well using the wireline logs and associated lithologies identified from cores in the same or nearby well. Initially, the persistence behaviour of the wireline logs is confirmed by rescaled range (R/S) analysis. These log data follow a local trend of lithological variation and hence can be used for lithology identification. Subsequently, we have used HCA and multi-layer feed forward (MLF) neural network to envisage the lithological sequence of varying thickness (thick beds of the order of 6 m; thin beds up to 0.3048 m) through the analysis of gamma-ray, bulk density (RHOB), neutron porosity (Φ), sonic transit time (Δt) and photoelectric factor downhole logs. The results using the log data from the Expedition 02 of Indian National Gas Hydrates Program (NGHP-Exp.-02) demonstrate that these non-traditional approaches are suitable for analysing formation lithologies where core data associated with discriminating finer beds are available. The HCA and MLF network-predicted lithologies, made in this study, match realistically with the core derived lithologies, demonstrating their efficacy. Thus, the approach is quite useful for providing quick and accurate information on subsurface lithologies.