<![CDATA[Shear-wave velocity (Vs) is an essential metric for subsurface characterization and CO₂ storage evaluation. However, Vs measurements are frequently unavailable in mature fields due to limited data acquisition. This research employs a machine-learning approach utilizing a Fully Connected Neural Network (FCNN) to predict Vs and Vp/Vs logs at a potential CO₂ injection site within a heterogeneous carbonate reservoir. Seismic elastic properties, particularly Acoustic Impedance (AI) and Vp/Vs, play a crucial role in assessing reservoir capacity by linking elastic responses to petrophysical properties such as porosity and water saturation. Conventional approaches, including the Castagna empirical relationship and Multiple Linear Regression (MLR), are commonly used for Vs estimation. Nevertheless, these methods often inadequately account for fluid-related effects. To address this limitation, this study examines two predictive approaches: (1) indirect Vp/Vs derived from predicted Vs, and (2) direct prediction of Vp/Vs prediction using a FCNN model. The findings indicate that direct Vp/Vs prediction demonstrate stronger correlation with observed data (R = 0.8023) and improved sensitivity to lithological and fluid variations compared to traditional methods. These findings underscore the advantage of directly predicting fluid-sensitive elastic properties through machine learning, providing a more reliable framework for reservoir characterization and CO₂ storage assessment in data-constrained carbonate formations.]]>
