Machine Learning-Based Pore Pressure Prediction in Deep Reservoirs: Model Comparison and Optimized Application of XGBoost

ABSTRACT: Considering the evolution process of regional sedimentary tectonics in deep formations, the causes of abnormal high pressure in each formation are different, and pore pressure prediction in deep formations remains a challenge in the oil and gas community. This paper studies the application of the optimized Artificial neural network, the support vector machine model, and the XGBoost (eXtreme Gradient Boosting) model for pore pressure prediction in target deep formations. It shows that different high-pressure mechanisms at different depths in deep formations dominantly lead to differences in the pore pressure field. When considering well log input parameters, four geophysical response logging parameters can achieve the best accuracy. Under the same training conditions, the XGBoost training model provides the best prediction results, followed by the support vector machine model and the Artificial neural network. The research results are very useful for pore pressure prediction in deep formations and improving the accuracy of the drilling mud density window. The findings are of great significance for advancing pore pressure prediction in deep formations and enhancing drilling safety.