Predicting the Elastic Properties of Anisotropic Reservoirs from Dipole Logging While Drilling Using Machine Learning

Abstract The presence of a drill collar in logging-while-drilling acoustic instruments complicates the calculation of subsurface formation properties from acoustic logs. The collar's acoustic modes interfere with the formation's acoustic modes, which are essential for determining rock properties, particularly the flexural mode obtained from a dipole source. This interference is more pronounced in slower formations. Furthermore, most formations exhibit anisotropy, yet conventional log interpretation typically captures elastic properties in only one direction. To address these challenges, a method is proposed that overcomes tool effects and accurately calculates the complete set of elastic properties of rocks from the dipole flexural mode. The approach utilizes the modified matrix pencil algorithm to extract the flexural mode with minimal noise and interference, followed by a neural network to calculate the anisotropic velocities of rocks. The results demonstrate that the predicted elastic properties are accurate, and that the method is robust and fast, enabling real-time prediction of formation properties during drilling. This application optimizes the drilling and geosteering of wells for hydrocarbon and geothermal production, as well as carbon sequestration applications. Real-time inference of reservoir properties reduces non-productive time, minimizes drilling risks, and enhances the ability to precisely reach target reservoirs. Overall, this approach improves the efficiency and effectiveness of subsurface characterization, contributing to more sustainable energy exploration and carbon management strategies.