Simplification and Simulation of Fracture Network Using Fast Marching Method and Spectral Clustering for Embedded Discrete Fracture Model

Abstract

Embedded Discrete-Fracture Model (EDFM) is designed to accurately represent realistic hydraulic fracture network (HFN) and provide efficient performance predictions by honoring the fracture topology. Due to the complexity of HFN, the EDFM grid may be computationally inefficient, particularly for field-scale applications with millions of fracture cells. This paper aims at incorporating the Fast Marching Method (FMM) and spectral clustering for fast HFN analysis, simplification and simulation under the framework of EDFM. HFNs are first generated using a commercial hydraulic fracture simulator. The FMM is used to solve the pressure front propagation using the fracture graph and subsequently the ‘diffusive time of flight’, well and completion index are calculated. The results are used as pre-conditions to split the fracture graph into connected components, which are subsequently partitioned using spectral clustering. The resulting clusters are used for fracture simplification resulting in a significantly lower number of fracture elements for flow simulation. To demonstrate the feasibility of the workflow, we use the Multi-Well Pad pilot model, which is characterized by a complex HFN and a high-resolution matrix system. We investigate the relationship between matrix resolution (characterized by the matrix-fracture size of the reservoir cells) and the ratio of oil and gas production on the field. Our investigation provides an alternative approach to explain the very large Gas Oil Ratio (GOR) reported for this type of reservoirs. The required levels of refinement to correctly represent the observed GOR presents an opportunity to test the efficiency and accuracy of our proposed workflow for HFN simplification. We use the results of the FMM applied to the high-resolution models to find an optimal spectral fracture clustering. The results show that the proposed workflow can achieve massive fracture cells aggregation (with only 1% of the original fracture cell number) while maintaining the accuracy. This is the first study for analysis, simplification, and simulation of HFN for EDFM using a field scale model. The main contributions are: (i) honor the topology of complex HFNs in EDFM and is able to represent the complex physics observed in the oil and gas shale reservoirs, (ii) HFNs diagnosis without simulation, and (iii) massive fracture aggregation with an error below 5 percent, and speed-up higher than 16 times of the fine scale model.