Abstract
Fractured reservoir simulation plays a crucial role in understanding various subsurface geo-energy recovery and storage processes, including shale gas/oil extraction, enhanced geothermal systems, and CO2 sequestration in basaltic rocks. However, such simulations often entail significant computational expenses due to the high contrast in permeability and pore volume (PV) between matrix and fractures. To address this challenge, we introduce a reduced-order model (ROM) tailored for fractured reservoir simulation that offers flexible fracture representations by generating coarse matrix nodes based on reservoir outlines and adding extra diagonal connections between unconnected matrix nodes, whose corresponding volumes are intersected by fractures. This approach avoids the need for additional fracture nodes, effectively reducing computational costs. Dimensionality reduction methods, such as principal component analysis (PCA), are used to give quality priors for sampling matrix transmissibility and PV arrays. Tuning to well observation data, such as flow rates and bottomhole pressures (BHPs), is achieved through a gradient-based optimization method within a general automatic-differentiable simulator framework. Our results demonstrate robust calibration using synthetic well observation data from a fine-scale reference simulation model. Incorporating dominant flow physics, such as water breakthrough, from observation data improves history-matching (HM) convergence and prediction accuracy. Additionally, PCA for parameterization enhances the convergence rate of model calibration compared with random initialization. Calibrated transmissibilities align with high-connectivity regions from the fine-scale reference model, rendering the model interpretable in terms of reservoir connectivity or geology. This feature enables our method to be used effectively for reservoir HM and optimization using field observation data.
