Combining classification and regression for improving shear wave velocity estimation from well logs data

A shear wave is a critical property that helps to constrain rock properties such as lithology, pore fluid, and pore pressure. Nevertheless, valid and reliable estimation of shear wave velocity in highly heterogeneous reservoirs is still considered a challenge to be faced. In this paper, we propose a strategy for improving shear wave velocity estimation based on the combination of clustering and classification plus regression using conventional logs. The approach includes two steps. In step one, data clustering algorithms are applied to build a classifier model for identifying the distinct electrofacies (EF) based on similarity of well log responses. In step two, a distinct and specialized regression model is selected from more sophisticated methods (empirical models, statistical regression, virtual intelligence, and rock physics modeling) for determining the final shear wave velocity profile. The effectiveness of this proposed strategy is validated over a dataset from offset wells and their respective measured shear wave velocity data. The final results revealed that the proposed strategy that combines data mining task of clustering and classification plus regression, led to more uniform and better predictive performance of shear wave velocity (R-2 = 0.97) in comparison with the use of stand-alone rock physics model (R-2 = 0.70), virtual intelligence (R-2 = 0.82), and statistical regression (R-2 = 0.91). The silent features of this proposed strategy is that field-specific regression models are built based on log derived EF and geological information.