Cost-effective parallel reservoir simulation on shared memory

Reservoir simulation is an important tool for petroleum engineers to predict oil production and optimize the management of oil fields. Nowadays, large-scaled reservoir simulations are required by oil industry to simulate complex geological models in order to obtain high resolution results. When the reservoir engineers design new production processes, it is necessary to run dozens of simulations to find optimal solutions. The speed of serial reservoir simulators can be a big challenge. Parallel reservoir simulators with fast computational methods should be invested. In this paper, we developed cost-effective parallel reservoir simulation techniques, which include mathematic fluid models, FIM discretization method, multilevel preconditioners and its implementations on shared memory. Based on the mathematical characteristics of the pressure and saturation in black-oil model, a multilevel preconditioner is set up which includes algebraic multigrid method (AMG), incomplete LU factorization, et al. The preconditioner is implemented on share memory to speed up the efficiency of numerical simulation. Several numerical experiments such as the benchmark test two-phase and three-phase SPE10 simulation and a field-scale mature waterflooding reservoir simulation were performed to test the efficiency, robustness of the proposed simulation technique and its parallel speedup. In the numerical experiments, the proposed reservoir simulation techniques can successfully simulate numerical cases with different reservoir properties on a desktop computer. Numerical results show that the proposed reservoir simulation techniques are quite efficient in solving the nonlinear PDEs and robust for highly heterogeneous field-scale problems.