Chapter 8 - Prediction of acoustic velocities using machine learning and rock physics

Abstract

Synthetic sonic logs are generated with a theoretical rock physics model (RPM) and are used to train three different machine learning algorithms for velocity prediction: support vector regression, random forest, and multi-layer perceptron. Random forest model best emulates the theoretical modeling of acoustic P-wave and S-wave velocities among the three models. For measured well logs from the Volve oil field in the North Sea, machine learning methods achieve significantly smaller prediction error than does the representative, but highly simplified, RPM that we used, with the best average coefficient of determination (R2 score) increasing by 8.0% for P-wave velocity prediction, and 63.9% for S-wave velocity prediction relative to the RPM. Finally, a proposed hybrid approach, combining machine learning and RPM, is found to improve prediction robustness, with the average R2 score increasing by 13.3% for P-wave velocity prediction relative to the RPM.