Pore Pressure Prediction Model Based on XGBOOST: A Case Study of Complex Geological Strata in Block K

ABSTRACT: Pore pressure stands as a pivotal element in ensuring wellbore stability and is instrumental in shaping the design of drilling fluids, safety protocols, drilling speed, and cost efficiency in the realm of oil and gas exploration and production. Traditional pore pressure prediction methods, including the Eaton method, equivalent depth method, and empirical coefficient method, often introduce notable uncertainties and biases due to challenges in accurately determining the in-situ stress coefficient, limited neighboring well data, and applicability across different formations. The intricate interplay of lithology and irregular faulting patterns further contributes to non-uniform distributions of pore pressure within reservoirs, complicating assessments using simplistic mechanistic formulas. However, recent strides in machine learning and data analytics have spurred the development of hybrid models that combine contemporary computational techniques with conventional methods, leading to substantial advancements in evaluating pore pressure within complex geological settings. This study showcases a groundbreaking model that integrates geological and logging data with a Shapley value-based interpreter to enhance prediction accuracy and the interpretability of engineering parameters. The results unequivocally demonstrate that the XGB model surpasses other methods in predictive performance and stability, with Shapley values offering invaluable insights into pressure variations. 1. INTRODUCTION Pore pressure is a crucial determinant in evaluating the stability of wellbore during drilling activities, impacting various operational aspects including drilling fluid design and safety protocols, as well as influencing mechanical drilling efficiency and cost management (Bourgoyne et al.,1986; Fan et al., 2015). The integrity of these predictions directly affects hydrocarbon exploration and the associated production processes. Traditional approaches to estimating pore pressure include methods such as the Eaton, equivalent depth, and empirical coefficient methods. However, these techniques often encounter significant challenges due to difficulties in obtaining continuous data regarding in-situ stress conditions, coupled with the limitations posed by seismic data resolution and the variability of rock mechanics parameters derived from well log interpretations and theoretical models (Lai et al., 2023). Such factors contribute to inconsistencies in the results produced by these theoretical models. Furthermore, the inherent lithological diversity and geological complexity further complicate the accurate forecasting of pore pressure distributions and variations (Li, 2024).