Developing robust machine learning techniques to predict oil recovery: A comprehensive field and experimental study

Abstract

The volatility in the oil industry driven by significant market demand and notable resource reduction, underscores the crucial requirement for developing a reliable and robust framework to promote oil recovery strategy. This study integrated various robust Machine Learning (ML) algorithms including the Decision Tree (DT), Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Regression (SVR), Gradient Boosting Regression (GBR), and Multilayer Perceptron (MLP) to predict oil recovery based on field and experimental data. Leveraging these models enhances prediction efficiency and reduces reliance on traditional methods. The performance of the integrated ML models with oilfield and experimental datasets, as well as the impact of multiple input parameters against traditional decline curve analysis (DCA) models, was evaluated. The findings reveal that RF, DT, and GBR models have achieved remarkable performance in contrast with other ML models and traditional DCA methods. The RF model has achieved the highest performance, reflected by a coefficient of determination (R²) value of 0.99 for both field Datasets (A) and experimental Datasets (B). More so, we accurately assess the ML model's robustness and performance by leveraging various metrics performance, including the Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC), to prove the robust alignment with a remarkable merit of accuracy and complexity across the integrated ML models. Ultimately, the results supported the RF model, which obtained the lowest AIC and BIC values among all the models for oil recovery prediction in Datasets (A) and (B).