The development of unconventional reservoirs require key decisions to be made under uncertainty. We regularly consider multiple variables, such as the number of wells to be placed in a section, their horizontal spacing and vertical staggering, the length and orientation of the laterals, the design of fracturing stages and associated perforations, the quantity and type of fracturing fluid and proppant to use, and geologic variability. These decisions are dependent on the subsurface parameters at each development due to the heterogeneities of rock and fluid properties as well as the nearby historical development. Such complex and dynamic problems combined with the fast pace and large scale of unconventional development pose a significant challenge for classical physics-based reservoir models. We propose a data-driven modeling methodology that is used to support development decisions in the STACK play in Oklahoma.
We utilize multi-variate analytics to model the behaviors of horizontal wells in the STACK (Sooner-Trend-Anadarko-Canadian-Kingfisher) of the Anadarko basin (see Fig. 1 for idealized geologic cross section). The STACK consists of two primary targets, the Meramec shale and the Woodford shale. The Mississippian age Meramec is 200-500’ thick with porosity ranging from 3-6%.The Mississippian/Devonian age Woodford ranges from 75-300’ thick with 3-7% porosity. The Meramec formation consists of several parasequences of fine-grained silts with significant carbonate input in some intervals. Our analysis includes over 500 horizontal wells that target the core intervals in the Meramec. We use these wells in our workflow that predict their production performance impact with respect to both the location of horizontal wells in the STACK trend and multiple engineering variables.
In unconventional resource plays, wells drilled and completed in the same area, within the same target and with the same completions can have results that vary by +/− 50% vs the mean. Without a predictive model to explain this variance, production variability looks like random noise. We chose to use a simple statistical technique called a hypothesis test, and interpret the result using a p-value, a measure of statistical significance. Using the p-value, a trend can be tested to see if the trend in a sample of data is statistically different than the trend that would be expected from a random sample. Several studies have been published on multivariate analytic workflows1,2,3,4.
Number of Pages
Looking for more?
Some of the OnePetro partner societies have developed subject- specific wikis that may help.