Prediction of shear wave velocity in the williston basin using big data analysis and robust machine learning algorithms

Abstract

The shear velocity is one of the most critical parameters in determining the mechanical rock elastic properties, which serve as inputs for different studies such as wellbore stability, mechanical earth modeling, hydraulic fracturing, and reservoir characterization. However, the sonic log is not acquired in every drilled well. We analyzed the log data of more than 35000 wells in the Williston Basin, and we found that only very few wells had sonic logs. For this reason, several studies attempted to correlate the shear velocity (or slowness) to other easily accessible properties; these will be presented in the literature review, with their pros and cons.

The focus of this paper is to apply machine learning algorithms to synthesize the shear slowness log. Our models are trained and tested with log data from 27 wells drilled in the Bakken petroleum system, Williston Basin. Logging data include Gamma Ray, Deep Resistivity, Density, Neutron Porosity, and Shear Slowness. Five different algorithms were developed and tested against blind data including Xtreme Gradient Booster, Random Forest Regressor, Linear Regression, Ada Boost Regression, and Bayesian Ridge Regression. Overall, the R2-score varied from 0.55 to 0.92, with the XGBoost outperforming the other algorithms