@Article{ee.2026.082858,
AUTHOR = {Qiantao Jiang, Shu Jiang, Wei Li},
TITLE = {Rock Mechanical Parameter Prediction for Deepwater Reservoir Using Different Machine Learning Algorithms},
JOURNAL = {Energy Engineering},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/energy/online/detail/27738},
ISSN = {1546-0118},
ABSTRACT = {Accurate acquisition of rock mechanical parameters is of critical theoretical guidance and engineering application value for key processes in deepwater oil and gas exploration and development, such as wellbore stability evaluation and fracturing stimulation design. Aiming at the problems of high cost, low efficiency, large prediction error and poor generalization of traditional core experiments and empirical formula methods, as well as the common missing of acoustic logging data in deepwater reservoirs, this study constructs a logging-data-driven prediction framework for acoustic slowness and rock mechanical parameters based on four machine learning algorithms, namely Random Forest (RF), Long Short-Term Memory (LSTM), Support Vector Regression (SVR) and eXtreme Gradient Boosting (XGBoost). With conventional logging data as input features, the models are trained, validated and optimized through standardized data preprocessing, multi-strategy feature selection and Bayesian hyperparameter tuning. Quantitative evaluation results show that the Bayesian-optimized XGBoost model achieves the optimal prediction performance, with a coefficient of determination R<sup>2</sup> = 0.95, mean absolute error MAE = 0.42 μs/m and mean squared error MSE = 0.50 for acoustic slowness prediction. Residual-based uncertainty analysis further verifies that the model residuals are approximately normally distributed around zero with low prediction uncertainty. The dynamic Young’s modulus and Poisson’s ratio calculated from the predicted acoustic slowness are validated by core measurements, with mean absolute errors of 3.2 GPa and 0.021 and relative errors of 6.8% and 8.9%, respectively, which meet industrial application accuracy requirements. Compared with traditional methods and other machine learning models, the proposed framework has higher prediction accuracy, stronger generalization ability and better engineering adaptability, providing a reliable and efficient technical approach for the high-precision continuous acquisition of rock mechanical parameters in deepwater complex reservoirs with missing acoustic logging data.},
DOI = {10.32604/ee.2026.082858}
}