As the core means to improve the oil recovery, the efficiency of reservoir water drive development is significantly affected by reservoir heterogeneity and multiphase fluid interaction. Traditional numerical simulation methods face problems such as complex physical model and low computational efficiency when dealing with large-scale three-dimensional models. In this paper, an efficient three-dimensional simulation method is proposed, which combines unstructured grid finite volume method (FVM) and Adaptive Mesh Refinement (AMR). The details of high gradient flow are captured by local encryption, and the overall calculation amount is controlled. At the algorithm level, a hybrid solution strategy combining geometric multigrid (GMG) preprocessing with adaptive Krylov subspace method is adopted, supplemented by Newton-trust region correction to enhance the nonlinear convergence stability, and the computational efficiency is optimized based on MPI+OpenMP hybrid parallel framework. The experiment is verified by SPE10 standard model and the actual model of BZ oilfield in Bohai Sea. The results show that the recovery rate of this method reaches 46.5% after 10 years of simulation, which is highly close to the high-precision fine grid reference solution (46.8%), with a water cut error of only 0.6% and a water drive front matching degree of 0.97. In terms of computational efficiency, the time step can reach 5 days (about 3 times of the traditional method), the number of nonlinear iterations and linear iterations are reduced to 3.1 and 19 respectively, and the memory occupation is reduced to 4.3TB, and the acceleration ratio is 3.78 in the parallel environment of 128-512 cores. In practical application, the deviation between the predicted value of oil increase and the actual production data of measures such as reconstruction of injection-production well pattern and optimization of layered water injection is small, which verifies its practical value in reservoir development decision. This study provides reliable technical support for remaining oil analysis and scheme optimization in water drive development.
F. Wang (Sun,) studied this question.