Optimizing Well Selection in Hydraulic Fracturing Using Advanced Machine Learning Approaches

Abstract

This research evaluates the performance of artificial neural networks (ANNs) and adaptive neuro-fuzzy inference systems (ANFIS) in selecting well candidates for hydraulic fracturing (HF) in the Bach Ho oilfield, Vietnam. Traditional well selection often depends on expert judgment and deterministic criteria, which may be limited in uncertain and data-constrained reservoir environments. To address this limitation, machine learning models are applied to improve decision-making accuracy. A dataset of 41 wells was analyzed using permeability, porosity, skin factor, reservoir pressure, water cut, and reservoir thickness to predict post-HF daily production rates. Both models were trained and evaluated using RMSE, MSE, MAE, and R². The ANFIS model demonstrated superior accuracy, achieving an RMSE of 4.24, R² of 0.93, and MAE of 4.24 on the training set. On the testing set, ANFIS achieved an RMSE of 40.44, R² of 0.81, and MAE of 30.33, outperforming the ANN model, which recorded an RMSE of 40.43, R² of 0.59, and MAE of 31.86. These results suggest that ANFIS is more effective in capturing nonlinear relationships and handling input uncertainties. The study presents a practical, interpretable tool for supporting petroleum engineers in prioritizing HF candidates, ultimately enhancing oil recovery and resource allocation in complex reservoir settings.