Application of XGBoost on Fast Prediction of Hydraulic Fracture Propagation

ABSTRACT: This paper presents a fast fracture propagation prediction method using the XGBoost algorithm, designed to significantly reduce computation time while maintaining high accuracy. By utilizing key parameters such as geomechanical properties, fracturing operation data, and natural fracture distributions, the method efficiently predicts fracture geometry and properties. Latin Hypercube Sampling (LHS) is employed to generate training data, and the model is trained on numerical simulations. Testing on real field cases shows the method predicts fracture geometry with a relative error of 0.6% in length and 1.9% in height for simple fractures, and under 5% for more complex multi-fracture cases. The method's prediction time is proportional to the number of time steps and remains constant regardless of the complexity of the case, requiring only 13 seconds for simple fractures and 85 seconds for more complex ones, compared to traditional simulations that take several minutes or hours. This makes the method highly suitable for real-time hydraulic fracturing operations, providing accurate and timely predictions to support decision-making in complex geological environments. 1. INTRODUCTION Hydraulic fracturing, commonly known as "fracking," is a critical technology for unlocking the vast potential of unconventional reservoirs, such as shale formations. Its importance stems from its ability to access and extract oil and natural gas resources that were previously unreachable. By injecting a high-pressure fluid mixture into rock formations, fractures are induced, creating pathways through which hydrocarbons can flow more easily toward the wellbore for extraction. The geometry of these fractures plays a key role in determining the efficiency of the fracturing operation and directly impacts the production performance of the reservoir (Li et al., 2019; Sun et al., 2021; Liu et al., 2023). Numerical simulations of hydraulic fracturing have become indispensable tools for predicting fracture geometry and understanding the mechanisms driving fracture propagation (Wu et al., 2015). Despite the widespread use of hydraulic fracturing simulations, one of the primary challenges lies in their computational efficiency, particularly in reservoirs with complex geomechanical properties and dense natural fracture networks (Li et al., 2020). These simulations require detailed models of fluid-rock interactions, fracture growth, and stress distribution, and the computational burden grows significantly with increasing geological complexity (Wu et al., 2014). In reservoirs with heterogeneous rock properties or intricate natural fracture systems, traditional simulation methods often struggle to accurately capture fracture behavior. This results in extended simulation times, reduced predictive accuracy, and delayed decision-making during field operations.