Stacked machine learning models for accurate estimation of shear and Stoneley wave transit times in DSI log

Accurate estimates of the shear and Stoneley wave transit times are important for seismic analysis, rock mechanics, and reservoir characterization. These parameters are typically obtained from dipole shear sonic imager (DSI) logs and are instrumental in determining the mechanical properties of formations. However, DSI log may contain inconsistent and missing data caused by various factors, such as salt layers and spike phenomenon, which can cause difficulties in analyzing and interpreting log data. This study addresses these challenges and estimates the shear and Stoneley wave transit times in DSI Log using machine learning methods and common logs, including computed gamma ray (CGR), bulk density (RHOB), and compressional wave transit time (DTC), as well as depth-based lithology of different layers. Data from two wells in a field in southern Iran were used. Outliers and noise were carefully removed to improve data quality, and data normalization methods were implemented to ensure data integrity. Then, invalid DTC values were corrected and used to predict DTS and DTST. Finally, missing and invalid DSI Log values were predicted using the final models. Eight distinct machine learning models, such as Random Forest (RF), Gradient Boosting (GB), Support Vector Regression (SVR), Multiple Linear Regression (MLR), Multivariate Polynomial Regression (MPR), CatBoost, LightGBM, and Artificial Neural Networks (ANN), were independently trained and evaluated. The results show that Random Forest best predicted DSI Log parameters among all models. This approach facilitates subsurface interpretation and evaluation and provides a strong foundation for improving reservoir management and future decision-making.