Abstract and Introduction

Advanced hydraulic fracturing technologies and increasing oil prices gave an impulse to shale oil resources development in the U.S. in the past decade. Recent drop in energy prices spurred technological improvements and research in understanding of the resource to enable efficient production in changing economic conditions. Exploring the variability in individual well production and identifying the relationships between well performance and major input factors, such as geology and completion practices are essential to

  1. describe technological progress and success;

  2. reveal areas for further improvements; and

  3. predict production for all locations of future wells.

The latter is particularly important if one wants to make projections about the future shale resource development.

In this study, we focus our attention on the Bakken shale play recognizing two producing horizons, Middle Bakken and Three Forks. We apply various statistical and machine learning techniques that give insight into the factors that drive productivity. The calibrated models help us extend that knowledge to estimate productivity in undrilled locations. First, we study current productivity trends, explore the existence of functional relationships between operator's completion practices, geologic rock properties, time and oil production. Then, machine learning techniques are used to investigate how data availability affects predictive power of the models. Finally, we decide on the most suitable model to make predictions about well productivity for all locations of future wells.

In our analysis

  1. we find evidence of spatial and temporal heterogeneity in the data, which indicates that analysis of the entire Bakken Formation for all time periods could lead to suboptimal results;

  2. we account for the presence of vertical variability of geologic resources, which, if neglected, could lead to erroneous conclusions;

  3. we use statistical methods such as model-based recursive partitioning to identify spatial and temporal productivity regions, which not only shed light on the interplay between oil production, geologic resources, and operators' decisions, but also aid us in conducting an economic analysis for creating potential future oil-production scenarios; and

  4. we use machine learning technique to predict productivity in undrilled locations.

This content is only available via PDF.
You can access this article if you purchase or spend a download.