Grid-Sensitivity Analysis and Comparison Between Unstructured Perpendicular Bisector and Structured Tartan/Local-Grid-Refinement Grids for Hydraulically Fractured Horizontal Wells in Eagle Ford Formation With Complicated Natural Fractures

In the context of modeling fractured horizontal wells, unstructured grids have been applied to generate simulation meshes for complex fracture networks. It is necessary to investigate how to choose an unstructured mesh to accurately simulate production performance. In this paper, a new unstructured gridding and discretization work flow is proposed to handle nonorthogonal and low-angle intersections of extensively clustered fractures with nonuniform apertures. The work flow is then validated with two models in terms of production behavior and central-processing-unit (CPU) performance: a synthetic model with one horizontal well and orthogonal intersected hydraulic fractures built by tartan grid, and a field-scale local-grid-refinement (LGR) model with three horizontal wells and irregular hydraulic fractures in a slightly dipping reservoir created by a commercial software plug-in. Good-quality matches are obtained between unstructured and structured grids in both pressure and production behavior. Sensitivity analysis of the meshing parameters suggests that refinement in the vicinity of fractures has improved both early and late production of a well, whereas background density has a dominant effect on the late production. Background-grid type and orientation have less influence as long as they have the same grid density. Fewer cells can be achieved by increasing reservoir-background size and size-progression ratio, replacing unstructured-background grids with structured grids, and reducing the complexity of the fracture networks without loss of the accuracy, resulting in improved CPU performance. This study applies unstructured grids to simulate multiple horizontal wells with complicated fracture networks, and provides detailed comparisons between unstructured and structured grids. Most importantly, it resolves the question regarding how to choose an appropriate mesh to yield both accurate results and high-quality CPU performance.