The Neuquén Basin in Argentina is rapidly progressing through the early appraisal stage of its development lifecycle. As an emerging "Super Basin," the Vaca Muerta play presents a unique opportunity to apply recently developed advanced machine learning (ML) early during the appraisal and full-scale development of the Neuquén Basin. The Vaca Muerta poses a challenge to development teams due to a smaller amount of well data available. With limited well production, every well provides important information that can be extracted by machine learning.
Shell's technical team collaborated with third party Novi Labs to address this challenge by using machine learning to predict new well performance and deepen insight into production drivers in the Vaca Muerta. The technical team employed a collaborative and iterative approach with third party Novi Labs to successfully build and validate machine learning models that have demonstrated accuracy to actual well production. In this effort, Shell technical experts created complete datasets, configured ML models, and generated ML-derived forecasts for multiple development scenarios. Several combinations of subsurface features and completions features were used in training multiple ML models.
This paper shares the collaborative and iterative process applied by the team, provides examples of accuracy of ML-generated forecasts compared to actual production, and summarizes useful key learnings to repeat and continue success using ML as a valuable tool for early appraisal development.
Although the Neuquén Basin has all the elements of a "Super Basin," when compared to other North American shale plays, there a fraction of producing wells across the Vaca Muerta's various fluid regimes. For comparison, there are <100 producing wells in the volatile oil window of the Vaca Muerta compared to over 6000 producing wells in the Eagle Ford field in the U.S. Within the Shell-operated concession black-oil acreage, the well control is even less. In addition to dynamic modeling and standard well performance analysis, to further accelerate learnings and de-risk the Vaca Muerta more rapidly, Shell is working with third-party Novi Labs to build data analytics models to study the impact of subsurface and well design parameters on reservoir productivity. Across the Vaca Muerta acreage, there are large variations in subsurface properties such as porosity, saturation, total organic carbon, maturity, clay content, pay-zone thickness, and fluid properties, in addition to well design parameters such as lateral length, stage spacing, number of perforation clusters per stage, and fluid & proppant loading.