Vectorization and Parallel Processing of Local Grid Refinement and Adaptive Implicit Schemes in a General Purpose Reservoir Simulator

Abstract

Abstract With less computer memory usage and a fewer number of calculations, local grid refinement and adaptive implicit schemes can effectively achieve similar accuracy as fully refined, fully implicit schemes. However, these methods become less appealing on a vector processor because they are not as susceptible to vectorization as the latter. This paper discusses the development of a vectorized, parallel-processed algorithm for the local grid refinement and adaptive implicit schemes in a general purpose reservoir simulator. The adaptive implicit scheme is considered as a subset of the dynamic, multilevel, mixed implicit, local grid refinement models. The algorithm was tested on an enlarged 12th SPE comparative solution problem. Results indicate a fivefold speed improvement due to vectorization. Compared to a fully implicit, fully refined model, the local grid model requires 63% less computer memory and 42% less computer time without notable loss of accuracy.