A Novel Stochastic CatBoost Based Shear Wave Velocity Prediction and Uncertainty Analysis in Sandstone Reservoir Using Multi-Seismic Attributes

Shear wave velocity ( $V_{s}$ ) is a crucial parameter in the characterization of any reservoir, directly influencing the understanding of rock mechanical properties and fluid content. Empirical relationship and Rock physics modelling struggles to accurately predict the elastic properties including Shear wave velocity in reservoirs. This limitation arises because these reservoirs often have intricate structures with diverse pore types. While shear wave velocity has been mostly predicted using wireline logging data, there is need to predict this velocity in the uncored and away from the logged area. Although seismic data driven models can overcome this challenge, these deterministic models for predicting $V_{s}$ often fail to capture the inherent uncertainties associated with subsurface formations. This study introduces a novel approach using stochastic CatBoost, an advanced machine learning technique, for Shear Wave Velocity prediction and uncertainty quantification in reservoirs. By leveraging the strengths of gradient boosting and stochastic modeling, our method provides more robust and reliable predictions of $V_{s}$ , accounting for the variability and uncertainty in geological data. The proposed approach is validated using a comprehensive dataset from a reservoir, revealing that the proposed model achieves both high accuracy and low uncertainty in predicting $V_{s}$ in sandstone reservoir. Additionally, analysis of 2D sections indicates that the internal regions exhibit lower uncertainty compared to their boundaries, as determined by various realizations of the stochastic CatBoost model. For comparison, a plain CatBoost and a Gaussian Process models are also employed, indicating superior capability of the proposed method. This highlights the potential of the model to serve as a robust tool for robust Shear Wave Velocity prediction in sandstone reservoir.