An improved projection-based embedded discrete fracture model (pEDFM) for multiphase flow in fractured reservoirs

The discrete fracture-matrix (DFM) approaches based on conforming grids become popular for modeling fractured reservoirs in the last decade. However, the application of conforming DFMs at field scale is limited due to its prohibitive computational cost. In recent years, embedded discrete fracture model (EDFM) has received considerable attention as a promising alternative. EDFM incorporates the effect of each fracture explicitly without requiring the simulation grid to conform to the fracture geometry. A compromise between accuracy and efficiency could be achieved in EDFM by enabling the use of standard corner-point grids for the background matrix domain. Although many works confirm the high accuracy of EDFM for the solutions of pressure and velocity field, very few results have been presented to examine its accuracy for the saturation solutions from multiphase flow problems. This paper shows that EDFM can induce large errors for multiphase displacement processes, due to its incapability to capture the proper flux split through a fracture. For the first time in the literature we present a systematic evaluation on the performances of EDFM for multiphase flow and provide a detailed analysis to illuminate when and why the method fails. The analysis motivates us to exploit the projection-based extension of EDFM (pEDFM) as an effective method to resolve the limitations associated with EDFM. pEDFM is recently developed by Tene et al. (2017) to address the issue of flow barriers, and is based on the introduction of extended fracture-matrix fluxes. Moreover, we make several improvements upon the original pEDFM method. A physical constraint on the preprocessing stage is proposed to overcome the limitation in a 'naive implementation' of pEDFM. A number of test cases with different fracture geometries are presented to benchmark the performances of the improved pEDFM method for multiphase flow. Grid convergence studies are conducted for different numerical schemes. The results show that improved pEDFM significantly outperforms the original EDFM method.