The task of identifying key production drivers in unconventional reservoirs remains challenging, even after decades of exploration and production in North America during which tens of thousands of horizontal unconventional wells have been drilled and completed. Tens to hundreds of variables, categorized as reservoir quality, well architecture, completion, stimulation, and production metrics, are involved and there are many different interrelationships among the variables to be considered. Further, formation evaluation is typically minimal and there are unknown variables in the system that can only be guessed at, ignored, or proxied.
The author’s team has combined Geographical Information Systems (GIS) analysis and multivariate analysis using boosted regression trees for improved data mining results as compared to univariate methods. The purpose of this lecture is to discuss key elements of data mining in unconventional reservoirs, in order to raise awareness of cutting-edge statistical tools and methods being brought to bear in the industry. The presentation will provide highlights of real world examples of data mining projects in three different shale plays.
If there were only one idea for audiences to take away from the lecture, it would be that exploiting unconventional reservoirs is a highly complex task with many moving parts and data mining is a needed tool to be applied to better understand the importance of specific well productivity drivers. Another way to say it is that the talk is intended to provide the audience with improved statistical methods for the “statistical” plays so that multi-million dollar decisions can be truly data-driven.