Development of Empirical Correlations for Pore Pressure Prediction from Well Logs Using Multiple Linear Regression

Accurate pore pressure prediction is essential for safe drilling operations and effective reservoir modeling, especially in regions where overpressure from disequilibrium compaction poses significant challenges. These challenges can lead to issues such as fluid loss, kicks, differential pipe sticking, heaving shale, and blowouts. Traditional methods often fail to capture the complex relationships between formation parameters and pore pressure. This study utilizes a machine learning (ML) approach to capture these intricate relationships, developing two empirical correlations for pore pressure prediction. The first correlation includes lithological information (sand and shale), and the second does not. Both correlations are derived from a linear regression model fitted to the well-log datasets. The data used in this research was obtained from three wells in the Northern Carnarvon Basin of Australia. It includes parameters such as sonic interval transit time, density, gamma-ray, depth, and well diameter. Wells 1 and 2 contributed approximately 22,038 data points, which were divided into 85% for model training and 15% for validation. The 8,860 data points from the Well 3 were used to test the model's accuracy. The results, as evaluated by statistical metrics like mean absolute relative error (MARE), mean relative error (MRE), and root mean square error (RMSE), show that the newly developed correlations perform better than existing ones. The first model, which includes lithological data, demonstrated promising accuracy with an RMSE of 352.208. The second model, which does not include lithological data, surpassed the first with an RMSE of 342.105. These developed correlations offer enhanced predictive capabilities and effectiveness, making them suitable for estimating pore pressure in real-world drilling scenarios.