Predic1ng Gas EUR in Shale Plays Using Machine Learning Methods: A Compara1ve Study of Marcellus, BarneH, and Haynesville Shales

Shale gas reservoirs, such as Marcellus, Barnett, and Haynesville, play a pivotal role in meeting global energy demands. Accurate prediction of Estimated Ultimate Recovery (EUR) at an early stage is crucial for optimizing reservoir development strategies. Current methods face limitations in accurately estimating the EUR, highlighting the need for advanced techniques. This study explores the application of machine learning (ML) Random Forest (RF) to predict gas EUR based on completion data and initial production. A comprehensive dataset comprising approximately 15,000 records from diverse shale plays was compiled. Key parameters, such as True Vertical Depth (TVD), stage spacing, proppant loading, injected slurry volumes, lateral length, breakdown pressure, number of stages, type of proppant, and additional formation properties, were considered. Corresponding production metrics, including first-month production, 90-days production, 6-months production, 12-months production, and EUR, were collected. This dataset formed the basis for training and validating the ML models. This analysis underscores the importance of incorporating initial gas production data into the input variables for precise EUR prediction. The results demonstrate that the inclusion of this crucial factor significantly enhances the accuracy of predictions across all shale plays. This study pioneers a systematic approach to predicting gas EUR in diverse shale plays using advanced ML techniques. The incorporation of initial gas production data emerges as a key determinant for accurate predictions, allowing for more informed decision-making in the early stages of well life. This research represents a significant step forward in optimizing shale gas reservoir development strategies through innovative and data-driven predictive modeling.