Machine learning for forecasting production in tight oil reservoirs: Application to the Bakken formation of Williston basin

Abstract

Despite the tremendous growth of oil production from tight reservoirs in the US, characterizing and forecasting these tight formations faces significant uncertainties and risks. Machine learning (ML) techniques have emerged as insightful complementary methods to physics-based unconventional reservoir simulations for predicting oil production due to complex reservoir properties that make traditional modeling challenging. This study applies ML approaches to predict oil production from the Bakken formation in the Williston basin and employs normalized production indices (NPIs) by normalizing cumulative oil production with time and completion parameters. Key parameters include surface coordinates, lateral perforated length, total base water volume and total proppant used in hydraulic fracturing, number of stimulated fracture stages, and water volume and proppant per stimulated fracture stage. Five scenarios were analyzed: raw cumulative oil and four NPIs. Feature importance analysis was conducted using SHAP, followed by prediction using random forest, XGBoost, and Multilayer perceptron. Results revealed that well location ranked among the top features, demonstrating superior production potential at the eastern Bakken formation. Total base water and water volume per fracture stage showed a contrasting trend, indicating that the water distribution strategy is critical for stage-level performance. Proppant variables displayed complex non-monotonic relationships with multiple optimal ranges. XGBoost outperformed other algorithms. Stage-normalized NPI₂ achieved optimal prediction accuracy with 37% improvement in R² compared to raw cumulative oil, while spacing-normalized NPI₃ offered the most practical implementation. These findings demonstrate that completion optimization strategies should align with specific production objectives, providing quantitative guidance for improved capital efficiency in unconventional oil development.