Three principal factors that influence shale gas productivity are the reservoir quality (geology), the fluid type (which affects drive mechanism), and the completion methodology (which is affected by geomechanics). Out of all the reservoir fluids, dry gas has the highest recovery factor. Within dry gas shale resources, fluid type is not as big a factor compared to other hydrocarbon types. With fluid type out of the equation, where does geology and geomechanics contribute in the assessment of type well (TW) areas? In most occurrences, the geology and geomechanics is unaccounted for and only the factors that influence or quantify productivity (EUR, Cumulative Production, etc.) are utilized to generate TW areas. This paper demonstrates a unique, innovative approach, combining well-known statistical learning methods with detailed geologic and geomechanical maps to derive "like" geologic areas, lending to common well productivity. This new approach, not only reinforces the necessity of geologic mapping work, but more importantly takes out the geologic bias, while statistically exploring the data for geologic similarities.

The approach outlined in this paper takes a slightly different approach from a previous study [1], which grouped wells into different type wells by productivity measures (e.g. flowing pressures, shut in pressures, condensate gas ratio, flow potential, and separator oil and gas gravity). This approach focuses on generating similar geologic areas built around grid-based geologic and geomechanical maps, and relating these common areas back to type well performance. The common approach taken by both studies is the application of statistical learning methods, specifically principal component analysis and hierarchical clustering, to explore and group observations into similar regions.

This study includes various datasets, which range from production data to detailed geologic data for a frontier gas shale play in North America, but is applicable to most unconventional gas shales. This approach utilizes unsupervised statistical learning methods and infers that we are not trying to be predictive, but rather explore what geologically is influencing the type well areas.

The results of this study show that unsupervised learning techniques applied in a meaningful way can be used to delineate type well areas. Actual production data (Gas Rate and EUR) was utilized to verify the type well areas that have been generated using this method. In conclusion, six type well areas were generated and ranked by production results. It is recommended by the authors to utilize this approach in any frontier shale gas basin, in order to better streamline a type well workflow.

You can access this article if you purchase or spend a download.