Multiscale reservoir simulation for fractured models involves considerable numerical complexity arising from the sharp contrasts between matrix and fracture rock properties and the need for refined grid resolutions to capture accurate flow dynamics. To tackle this challenge, we develop a physics-based coarsening strategy that combines strongly coupled interpolation between coarse-scale and fine-scale solutions across matrix and fracture grid connections. Specifically, we introduce novel physics-based coarsening strategies that adapt the sizes of the coarse grid blocks in the matrix and fracture domains separately. Additionally, we enforce decoupling multiscale basis function calculations across matrix and fracture grid connections for general fractured models, significantly improving linear solver computational efficiency.

First, we observe that a discrete fracture model using unstructured grids exhibits slow linear solver convergence performance from the previous approach proposed by Watanabe et al. (2023). Numerical experiments reveal inaccurate basis function interpolation from coarse-scale to fine-scale pressure solutions along fracture networks due to insufficient support region overlapping across fracture coarse grid blocks for basis function calculations. Additionally, we visualize the matrix and fracture fully coupled coarse-scale pressure matrix sparsity pattern highlighting a lack of diagonal dominance. The developed strategies ensure smaller coarsening along fracture networks while allowing larger coarsening in matrix domains, effectively reducing the coarse-scale pressure matrix size. Consequently, this strategy accelerates overall linear solver convergence by minimizing the interpolation errors in the multiscale preconditioning stage.

We rigorously evaluate the developed algorithm performance on both discrete fracture and embedded fracture models in terms of linear and nonlinear solver iterations and central processing unit (CPU) times. Additionally, we benchmark the new multiscale pressure algorithm performance against a standard fine-scale pressure solver. In summary, our approach demonstrates accurate and efficient field-scale fracture modeling in multiscale reservoir simulations.