A facies-constrained flow-network model for fast history matching and production optimization of polymer flooding reservoirs

With the rising water cut in mature oil fields, polymer flooding has emerged as a critical Enhanced Oil Recovery (EOR) technique. However, high-fidelity numerical simulations for history matching and polymer flooding optimization remain computationally intensive, limiting their practicality for Closed-Loop Reservoir Management (CLRM), which is inherently dependent on rapid iterative simulations for real-time model updating and operational decision-making. Although physics-based data-driven flow-network models, such as General-Purpose Simulator-powered Network model (GPSNet), can accelerate simulations, their lack of geological constraints compromises predictive reliability. To address this limitation, we propose a novel facies-constrained flow-network model (GPSNet-FC) within the GPSNet framework. This model simplifies reservoir geometry into a 1D discretized grid between wells while incorporating sedimentary facies boundaries identified through edge detection and level-set methods. Grid properties are assigned and calibrated based on facies-specific attributes to ensure geological consistency. GPSNet-FC is applied to history matching using the Ensemble Smoother with Multiple Data Assimilation (ESMDA) and to polymer flooding optimization via the Differential Evolution (DE) algorithm. Numerical case studies validate the method, demonstrating that GPSNet-FC outperforms the original GPSNet in both reliability and accuracy. By integrating facies-based geological constraints, this approach reduces non-uniqueness in history matching and enables rapid and accurate decision-making for polymer flooding strategies. This work advances the integration of geological data into physics-based data-driven models, offering a robust and efficient tool for the CLRM of polymer flooding reservoirs.