Predictive Modeling of Petrophysical Properties in Lake-Facies Carbonates: A Case Study of Qaidam Basin

This study investigates the petrophysical attributes of the Nanyinshan reservoir situated in the Qaidam Basin, an area predominantly characterized by lake-facies carbonate rocks, emphasizing the predictive modeling of porosity using machine learning techniques. The geological analysis reveals the complex depositional systems and tectonic influences shaping the reservoir, with data from the Yi Ping 1 well serving as the foundation for this research. This area is predominantly composed of lake-facies carbonate rocks. A comparative evaluation of six machine learning algorithms, including Random Forest (RF), XGBoost (XGB), Support Vector Regression (SVR), Ridge Regression (RR), Polynomial Regression (PR), and Linear Regression (LR), were conducted to predict porosity. Among these, RF emerged as the most effective model, achieving the highest predictive accuracy (R2 = 0.9046) and the lowest error metrics (Mean Squared Error(MSE) = 0.93, Root Mean Squared Error(RMSE) = 0.97, Mean Absolute Error(MAE) = 0.81). Residual error analysis highlighted RF's superior consistency, characterized by a narrow interquartile range and minimal bias, while simpler models like LR struggled with significant variability and sensitivity to outliers. Advanced models, such as Random Forest (RF) and XGBoost (XGB), demonstrated a robust ability to capture the non-linear relationships inherent in the dataset, outperforming traditional approaches. The results lay the groundwork for the integration of predictive analytics into reservoir management practices, mitigating uncertainties in resource assessment and enhancing exploration and production strategies. Subsequent research will aim to integrate further geological and geophysical datasets to enhance model accuracy and applicability.