The application of EEI inversion based on shear velocity prediction from random forest

Abstract This paper introduces the random forest algorithm, a machine learning method, for shear wave velocity (Vs) prediction. The methodology involves the following steps: First, basic well log curves and measured shear wave velocity curves from an oilfield are collected. The input datasets are then randomly split into two subsets—a training set (80%) and a testing set (20%). Initially, separate machine learning models are constructed using the training and testing datasets to establish relationships between the shear wave velocity curve and the input log curves. Subsequently, the model’s reliability is assessed by evaluating the coefficient of determination (R 2 ) for both subsets. If the model proves reliable, it is applied to predict shear wave velocities in target wells where such measurements are unavailable. Once the Vs predictions are obtained, Extended Elastic Impedance (EEI) inversion can be performed to enhance reservoir characterization. A case study from an offshore oilfield demonstrates that the random forest-based shear wave velocity prediction method is effective for quantitative seismic interpretation.