P-Wave Velocity Log Simulation for Petroelastic Inversion Using Gardner’s Equation, Neural Networks And Ensemble Methods Based on Trees: A Case Study of the Santos Basin, Brazil

Abstract

 One of the main tools for reservoir characterization is analyzing well-log data. The importance of such methods stems from petrophysical properties estimation, such as porosity, which is very important to the oil and gas industry. In scenarios where data is hard to collect, data loss and technical failures during the acquisition impose an extra challenge. Thus mathematical and petrophysical models are good candidates to fill information gaps in the well-log dataset. In such a way, the rock’s petroelastic and petrophysical properties can be successfully estimated. Several studies correlate the velocity of compressional waves (VP) to other basic well data. In this study, we used the Gardner equation and Machine Learning methods such as Neural Networks, Random Forest and Gradient Boosting regressions to generate VPlogs. We used real-world data acquired from twenty wells of the pre-salt formation from Santos Basin in Brazil to train and test the Machine Learning methods and evaluated the data estimated by those models using statistical metrics. We calculated the acoustic impedance from the estimated logs and used it to create a prior model for a petroelastic inversion, which allowed us to estimate the natural logarithm of the acoustic impedance for a seismic volume. The Machine Learning methods presented lesser errors between estimated and measured velocities when compared to Gardner’s equation.