Dual–Continuum Coarse–Grid Network: A Hybrid Approach for Naturally Fractured Reservoir Simulation

Abstract Reservoir simulation for naturally fractured reservoirs pose significant challenges due to complex flow behavior driven by fracture–matrix interactions, early fluid breakthrough, and strong heterogeneity. While full-physics fractured reservoir simulation models like the Embedded Discrete Fracture Model (EDFM) offer high fidelity, its relatively high computational cost still limits the applications in iterative workflows like reservoir history matching and optimization. This study introduces the Dual-Continuum Coarse Grid Network (DC-CGNet), a hybrid physics-based data-driven model designed for efficient simulation and history-matching of fractured reservoirs. DC-CGNet employs a coarse dual-layer graph, where matrix and fracture are represented as overlapped graphs. The model parameters in DC-CGNet include relative permeability curves, transmissibility, pore volumes, and the well indices. Due to its coarse discretization nature, the number of model parameters in DC-CGNet is significantly lower than EDFM or traditional Dual-Porosity Models (DPM). The history matching of DC-CGNet can be achieved through coupling with the Ensemble Smoother with Multiple Data Assimilation (ES-MDA). DC-CGNet was examined through a synthetic water-flooding case. The results show that DC-CGNet can accurately history match and forecast dynamic well responses, and achieve up to 15 times speedup. The proposed framework offers a practical, scalable solution for fractured reservoir modeling and history matching, and can be readily integrated into existing reservoir simulators to help guide fast decision-making in reservoir management.