An Adaptive Preconditioning Strategy to Speed up Parallel Reservoir Simulations

Wang, Kun (University of Calgary) | Liu, Hui (University of Calgary) | Yan, Lin (Exploration and Development Research Institute, PetroChina) | Luo, Jia (University of Calgary) | Wu, Keliu (China University of Petroleum) | Li, Jing (University of Calgary) | Chen, Fuli (Exploration and Development Research Institute, PetroChina) | Dong, Xiaohu (China University of Petroleum) | Chen, Zhangxin (University of Calgary)


In reservoir simulation, an ILU preconditioner is the most widely used preconditioner for preconditioning linear systems due to its simplicity and low computational cost. However, an ILU preconditioner sometimes is not effective enough, especially for a large-scale parallel reservoir simulation problem with a highly heterogeneous geological model. A constrained pressure residual (CPR) preconditioner is considered a more efficient one, which employs two stages of a preconditioning process: the first stage uses the Algebraic Multi-grid (AMG) method to solve a pressure system, and the second stage uses the ILU method to solve the whole system. Its disadvantage is the high computational cost of the AMG method. In order to reduce the computation costs on preconditioners and the resulting linear solvers, we have developed an adaptive preconditioning strategy [16] to automatically select a preconditioner between an ILU preconditioner and a CPR preconditioner or switch the ILU preconditioner to the CPR preconditioner and vice versa during a linear solution process. In this paper, the adaptive strategy is further analyzed and studied to understand its numerical performance and to choose optimal switch criteria.