Demystifying Child Well Degradation with Novel Feature Engineering: Part2 - Machine Learning Analysis for Child Well Degradation with Multiple Parent Wells in the Delaware Basin

Abstract As unconventional oil and gas development evolves, operators are increasingly drilling longer laterals—often exceeding 12,500 to 15,000 feet—resulting in more intricate parent-child well configurations. In maturing fields, child wells frequently transition into parent wells, creating scenarios where a single child well may be influenced by multiple parent wells developed at different times. This study builds upon our previous work (Ben et al., 2025), which focused on single-parent-child relationships, by extending the analysis to approximately 400 child wells with multiple parent wells across three core business areas. In this study, we introduced two novel engineered features: the Effective Parent Depletion Index (EPDI) and the Effective Parent-to-Child Distance. EPDI quantifies the depletion impact from multiple parent wells at the time of child well fracturing. It integrates four key factors for each parent well: (1) cumulative production, (2) lateral length, (3) diagonal distance to the child well, and (4) the degree of lateral overlap with the child well. These factors are combined into a single depletion index. The Effective Parent-to-Child Distance incorporates both diagonal distances and lateral overlaps. The two engineered features yield a more accurate representation of spatial interaction and potential interference between wells than those used in the ML model developed previously for the same asset (Wolfram et al, 2022). These features, along with the log-normalized first-year cumulative oil production, child-to-infill diagonal distance, weighted overlaps, and several geologically informed variables, serve as inputs to a machine learning (ML) model designed to predict child well performance. Monotonic constraints are applied to ensure that predicted production increases with greater parent-to-child distance and lower EPDI, aligning with physical expectations. A Shapley value analysis reveals that EPDI and effective distance are the most influential predictors of child well degradation. The model enables calculation of a degradation factor, offering a quantitative measure of parent-induced performance loss. The ML model has been deployed to the cloud, providing engineers with a practical tool for reserve estimation and infill planning. When integrated with an economic optimizer, this approach has the potential to unlock millions of dollars in additional value. The study highlights how domain-guided feature engineering enhances model interpretability and performance, offering actionable insights for optimizing well placement in complex development scenarios.