Traditionally, oil and gas asset performance in unconventional plays has been predicted using type curves derived from average well performance within sub-asset areas. This work explores operator applications of open-source machine learning algorithms in conjunction with customized workflows to develop an improved, objective approach to outlining type curve areas. Recognizing that well performance is ultimately driven by rock quality, we implement a data-driven approach utilizing basin-scale geological data to delineate areas of comparable geology using various clustering techniques and a wide range of parameter and hyperparameter exploration. Individual clustering outcome realizations are then high-graded using standard evaluation procedures and quantitatively ranked through a custom-built ranking algorithm utilizing well production data. Those cluster realizations that are most highly ranked are then reviewed and refined by business unit petrotechnical experts before final implementation to define type well profile analog sets and create geologically representative type curves. The approach outlined in this work provides an efficient, objective, and interdisciplinary methodology to derive geologically based type well profile areas that have resulted in direct improvements to business decision making through reduced type curve uncertainty.
Unconventional oil and gas assets present unique challenges in terms of well performance prediction due to the variability of subsurface conditions. To address these challenges, operators have typically employed type curve areas, i.e., subdivisions of the asset area wherein previously drilled and producing wells exhibit comparable performance (Chaudhary and Lee, 2017; Xiong et al., 2018). Once delineated, these type curve areas can be used to generate distributions of well production performance based on the wells contained within those areas. The distributions are then utilized to predict the performance and potential range of outcomes for newly drilled and completed wells.
Typically, type curve areas have been delineated directly on the basis of production data by taking the bounding geometry of wells gathered together on the basis of comparable production performance or geographical area (Bate et al., 2013; Jha and Lee, 2022). However, this approach suffers several notable limitations: 1) high-quality production data can be sparse, compared with other data sources (e.g., log data for geologic maps), 2) the production data that are available are often spatially biased because drilling activities are almost always focused in particular areas of any given asset, and, most importantly, 3) production data are complicated by a multitude of engineering factors such as differing completion designs, well spacing, depletion/parent-child effects, facilities/development strategies, and landing benches, which can inhibit the identification of representative performance trends.
