An Improved CPO-XGBoost Model for CO₂-WAG Development Parameter Optimization and Production Forecasting in Low-Permeability Reservoirs

CO₂-WAG (Water-Alternating-Gas) injection is a vital technology for mitigating viscous fingering and gas channeling in CO₂ flooding processes, with the rational design of injection parameters being essential for low-permeability reservoirs. However, the significant heterogeneity and complex geological characteristics of these reservoirs introduce challenges in optimizing development parameters and accurately predicting production performance. This study developed 1,225 multivariate numerical simulation cases for CO₂-WAG and utilized XGBoost as the foundational machine learning framework. To optimize the model, four metaheuristic algorithms—Crowned Porcupine Optimization (CPO), Grey Wolf Optimizer (GWO), Artificial Hummingbird Algorithm (AHA), and Black Kite Algorithm (BKA)—were applied. Among these, CPO demonstrated superior performance in handling high-dimensional, complex problems due to its balanced global search and local exploitation capabilities. Further improvements were achieved by integrating Chebyshev chaotic mapping and an elite opposition-based learning (EOBL) strategy, leading to the development of the ICPO (Improved Crowned Porcupine Optimization)-XGBoost model. The ICPO-XGBoost model achieved exceptional predictive performance, with a coefficient of determination R² of 0.9894 and a root mean square error (RMSE) of 2.894, while maintaining prediction errors within 2%. These results validate the model’s stability, reliability, and generalization capabilities, offering a robust and innovative solution for CO₂-WAG parameter optimization in low-permeability reservoirs.