Machine learning-informed ensemble framework for evaluating shale gas production potential: Case study in the Marcellus Shale

Artificial intelligence and machine learning (ML) are being applied to many oil and gas (O&G) applications and seen as novel techniques that may facilitate efficiency gains in exploration and production operations. Significant improvements in that regard are likely to occur when ML can be applied to evaluate O&G challenges with inherent synergies that may have otherwise not been evaluated concurrently. This study introduces an ensembled framework that couples a data-driven ML predictive model capable estimating a productivity indicator for unconventional O&G horizontal wells that correlates to estimated ultimate recovery (EUR) with a well design optimization approach that maximizes productivity. The framework is then applied to spatially rank productivity potential from low to high across the Marcellus Shale. The ML model developed used a gradient boosted regression tree (GBRT) algorithm and is capable of 82 percent prediction accuracy on holdout data. The distribution of geological properties as well as the resulting optimized well design and completion attributes specific to regions commonly ranked in productivity potential are evaluated statistically to comprehend controlling factors on shale well production, and to identify if commonality or disparity exists in the prominent features. The highest productivity ranked region is isolated in the Marcellus Shale's northeastern core region and its periphery. Statistical analyses indicate that regions higher in productivity ranking show a significant difference for certain (but not all) geologic features favorable to gas production potential relative to lower productivity regions; most notably net thickness and porosity. Optimized well design parameter settings vary relative to their placement across the study area and subsequent productivity ranking region. Overall, the ML-based framework discussed in this paper attempts to analyze shale controlling factors concurrently, to deliver a systematic evaluation result for production potential that accounts for and quantifies controlling features associated with geologic properties and well design attributes.