Answers are grounded in this paper and cite their sources.
Demonstrates enhanced oil recovery optimization in heterogeneous reservoirs, indicating improved efficiency.
Polymer flooding is a widely used enhanced oil recovery (EOR) method, but its efficiency is often limited in heterogeneous reservoirs due to uneven sweep and water channeling. This study presents a Physics-Informed Neural Network (PINN) framework to predict and optimize polymer flooding in a 2D heterogeneous reservoir, integrating physical laws with production data for accurate, real-time forecasting. The PINN model successfully captured pressure buildup, with values increasing from 200 to 212 bar near the injector and decreasing to 182–185 bar near the producer, while polymer front propagation reached 420–480 m in high-permeability zones and 250–300 m in low-permeability zones, with peak concentrations near the injector of 0.95–1.0 kg/m³. Cumulative oil recovery improved from 52% OOIP in the base case to 60–67% OOIP with PINN-based optimization, representing a 12–15% incremental increase. Model validation across training splits (70/30, 80/20, 90/10) showed close agreement with numerical simulations, with Mean Absolute Errors (MAE) of 0.48, 0.42, and 0.40 bar, and Root Mean Square Error (RMSE) values of 0.63, 0.58, and 0.55 bar, confirming the model’s robustness and generalization. The study demonstrates that the PINN framework accurately predicts reservoir behavior, captures heterogeneity effects, and enables real-time injection optimization, while reducing computational time by ~70% compared to conventional simulators. These findings showcased PINNs as a reliable tool for real-time tool optimization in oil recovery and operational decision-making in heterogeneous reservoirs.
No takes yet. Share an insight, caveat, or question.
Adaoma et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: