Prediction of Sonic Transit Time Using Machine Learning Techniques: A Case Study of the Oued Zine Field

Abstract

Sonic transit time (DT) is considered one of the most important petrophysical parameters used for reservoir characterization and subsurface formation evaluation. This study aims to predict DT values in the Oued Zine Field, located within the Sbaa Basin in the western Algerian Sahara, using artificial intelligence techniques based on both drilling data and logging data. Predictive models were developed using three machine learning algorithms: XGBoost, Random Forest, and Support Vector Machine to evaluate their capability in accurately estimating DT values. The results demonstrated the superiority of the XGBoost model compared to the other algorithms in both scenarios. Using drilling data as input variables, the XGBoost model achieved an R2 value of 0.9249 with RMSE and MAE values of 1.4324 and 1.0584, respectively, on the testing dataset. In contrast, when logging data were used as inputs, the model achieved an R2 value of 0.9105 with RMSE = 1.27658 and MAE = 0.955662, indicating a high predictive capability for DT estimation using both datasets. The findings confirm the effectiveness of machine learning techniques in improving reservoir characterization and reducing dependence on costly conventional measurements, while also highlighting the importance of drilling data for real-time prediction during drilling operations.