Unconventional Inflow Performance Relationship and Machine Learning Study

Abstract Unconventional wells, characterized by steep production decline, exhibit unique inflow performance relationships (IPRs) governed by transient reservoir behavior. A key challenge in modeling these wells is accurately estimating the static bottomhole pressure (SBHP). To address this, a hybrid methodology has been developed that combines physics-based reservoir modeling with machine learning (ML) to efficiently generate IPR curves for both historical analysis and future forecasting. The physics-based component involves calculating flowing bottomhole pressure (FBHP) using multiphase flow correlations, along with numerical simulations based on rate-transient analysis (RTA). These simulations were constructed for a representative subset of wells in the Permian Basin and calibrated using production history and shut-in (SI) pressure buildup data. Using these RTA models, 24-hour shut-in tests were simulated at various stages of a well's life to estimate pressure buildup (DP). A 24-hour shut-in pressure is adopted as a practical proxy for SBHP—supported by analysis showing that such durations yield sufficiently accurate IPR static pressure estimates in unconventional reservoirs. The simulated DP trends were then used to train a machine learning model capable of estimating DP, SBHP, productivity index (PI), and gas-oil ratio (GOR) for all wells across the Permian. The model incorporates reservoir characteristics, PVT data, and operational parameters, enabling scalable, data-driven IPR generation across a broad range of unconventional assets. This approach has been successfully deployed within Chevron's Digital Oil Field (DOF) operations. Unlike conventional reservoirs, where IPR is governed by pseudo-steady state flow and drainage area, unconventional reservoirs are dominated by stimulated rock volume (SRV), resulting in steep IPR decline. The workflow was initially applied to 400 wells in Delaware Basin across different benches. It was further scaled to the entire Chevron's Permian Basin assets. The ML models automate IPR generation for both lookback and forecast periods, extending up to 10 years. FBHP predictions by ML leverage downhole gauge data when available, or alternatively use wellhead pressure (WHP), wellbore geometry, flow path, and production rates. The SI DP model highlights the influence of current production, reservoir depletion, and completion design. Future productivity index (PI) is derived by integrating SI DP with production decline curve analysis (DCA). An IPR dashboard as a delivery from this initiative has been deployed across Chevron's unconventional portfolio, supported by automated big data preprocessing, continuous integration and continuous delivery pipelines, and the Azure ecosystem. This scalable solution enables robust IPR validation and application, reducing reliance on arbitrary SBHP and PI estimates in nodal analysis. This study presents a practical framework for deriving transient IPR in unconventional wells using integrated physical and ML modeling. The IPR outputs are embedded in the DOF to optimize production by aligning inflow and outflow performance. Deployment across Chevron's Permian unconventional inventory has contributed to an estimated 1.2% uplift in 2024 production.