[URTeC 2022] Accelerating field optimization for Shell in the Neuquén Basin using Novi Labs machine learning

Abstract::

Objectives/Scope: The Neuquén Basin is emerging as a “Super Basin” for unconventional oil and gas development; however, development teams for the Neuquén Basin are faced with the common appraisal challenge of sparse well control. Shell is working with third-party Novi Labs to apply machine learning models and data analytics in combination with physics-based modeling and standard well performance analysis to de-risk and optimize development in the Vaca Muerta. The machine learning techniques identify the co-dependent impacts of large variations in subsurface properties with spacing and completion designs to predict actual well production. The complex interaction of the variables on shale well production provides a key piece of analysis that increases confidence in full-field optimization.

Methods/Procedures/Process: We applied data analytics to machine learning models to generate scaling factors that tie multiple co-dependent variables directly to individual well production. The models were trained with a dozen subsurface features including porosity, saturation, maturity, and several others, combined with well spacing and completion parameters to target oil, water, and gas production rates. Completion design parameters included lateral length, fluid and proppant intensity, and stage spacing. The production predictions from the machine learning model were validated against actual production using established data analytics practices and a 80% train to 20% test split. Additionally, the machine learning model identified production drivers and analog wells to explain how it predicted well production.

Results/Observations/Conclusions: The scaling factors generated in this study provided valuable insight into the relative impact of multiple variables that co-dependently predict well production. These scaling factors informed the asset type-curve for future well inventory and facilitated optimization based on data instead of expensive experimentation using drilling and completion capital. In addition to the production predictions, the production drivers analogue wells identified by the machine learning workflow enabled the Asset Development Team to understand how the scaling factors were generated and qualitatively validate the accuracy of the predictions. The explainable model results enabled the asset team to build internal confidence in the machine learning models.

Applications/Significance/Novelty: The potential value of full-field optimization insight early in the life of the Neuquén Basin is immense. Machine learning data analytics models have demonstrated they can provide this information in other established North American Shale Plays, but this application in the Vaca Muerta acreage is the first opportunity to apply these techniques at scale at this early stage of field life. The data analytics model outputs inform asset type-curves, enable well and pad level optimization, and can eventually inform business planning and initial resource estimation. Finally, the scaling factors generated in this study illuminate a path where Shell can match full-scale development plans in the emerging Neuquén Basin to strategic economic objectives within Shell’s wider portfolio.

Interdisciplinarity: This interdisciplinarity work includes contributions from reservoir engineering, geosciences, well engineering, data science, analytics, and software engineering.