While the Wolfcamp A formation is considered a primary target in the Delaware Basin (as measured by activity and total production volumes), the Wolfcamp C has been steadily but less aggressively tapped by Permian operators for nearly a decade. In 2021, we worked to characterize this under-developed formation for potential inclusion into long range asset planning.
Defining the optimal development plan for the Wolfcamp C (WFMP C) in the northern Delaware Basin required a multidisciplinary effort that integrated geochemistry, geology, geophysics, and reservoir engineering data. This paper outlines two methods used to normalize large ranges of rock and fluid properties (spanning hundreds of square miles) for improved predictions of well performance. The "Analog Approach" requires an arbitrary selection of combined rock and fluid properties using "binary" cut-offs or risk bins. The Integrated Machine Learning (ML) Approach uses supervised ML algorithms to establish a similarity classification for user selected rock and fluid properties.
The Delaware Basin team leveraged two distinct approaches to define geologically similar areas (GSAs). The first approach utilized manual integration of log derived facies, petrophysics, seismic attributes and fluid properties. The second technique applied grid-to-grid unsupervised machine learning (ML) algorithms across geologic and geophysical maps. Reservoir engineering techniques were then used to quantify production interference as a function of wells per section, completion parameters, and confinement inside of each GSA. These results were then validated via integrating multiple data sources such as time lapse geochemistry (TLG), downhole pressure gauge data, rate transient analysis (RTA), microseismic, etc. Finally, once type wells were defined, economic engines and strategic plans determined the optimal development for this resource play.
This paper expands the workflow that Kreman et al., (2018) proposed, developing type wells across the basin that capture the regional performance variability. The integrated machine learning approach produced results that were consistent with the analog approach, but less labor-intensive and leveraged digital technologies to optimize field development.
Given the complexity of unconventional plays, it is critical to incorporate learnings from various data sources. The proposed integrated workflow helps constrain the uncertainty parameters to allow informed business decisions and optimized full field plan of development.
Workflows developed as part of the second approach (proposed ML generated GSAs) can be used for most phases of development. Because the data analytics driven approach matched well with our manual approach, this method can be automated to constantly evolve with additional data as constraints.