The Application of Unstructured-Gridding Techniques for Full-Field Simulation of a Giant Carbonate Reservoir Developed With Long Horizontal Wells

Abstract

Summary This paper presents a gridding study relating to reservoir simulation of a giant, complex, low-permeability carbonate reservoir developed with 75 ultraong horizontal wells in a densely spaced alternating injector/producer pattern. The lateral magnitude of the Al Shaheen field in Qatar and the radial layout of the multiple ultralong horizontal wells in the field posed a challenge in modeling of individual well performance using a manageable grid size with an acceptable run time for history matching. Reservoir modeling was complicated further by the complex reservoir characteristics with a tilting free-water level (FWL), separate gas caps, large lateral variations in oil properties, and wettability-dependent flow characteristics. These features had to be incorporated into the initialization and dynamic modeling of the reservoir, which added further to the memory requirements of the simulation model. This paper describes the process of selecting a suitable simulation grid for history matching the performance of this reservoir on a full-field basis. Conventional Cartesian gridding techniques, including the use of local grid refinements (LGRs) in areas of interest, were pursued initially but were shown to be inadequate for full-field modeling of this complex reservoir. The gridding problem was solved by the use of 2.5D perpendicular-bisector (PEBI) grids around each of the horizontal wells in the field. This allowed for sufficient resolution between wells and also aligned the grid with the well paths, thereby avoiding grid nonorthogonality issues. The efficiency of the PEBI model was also demonstrated by the comparison of CPU performances. Run times for the full-field PEBI model were equivalent to that of a conventional Cartesian model with suitable local grids covering only 20% of the wells. Both models had approximately 700,000 active cells and required 3-4 GB of memory. A full-field model relying on conventional LGRs around all wells was not built because it would involve significantly more grid cells and, therefore, would become considerably slower and require more memory.