An intelligent multi-fidelity surrogate-assisted multi-objective reservoir production optimization method based on transfer stacking

Recently, many researchers have focused on reservoir production optimization because it is one of the most essential processes in closed-loop reservoir management. Surrogate-assisted production optimization in particular has received a lot of research attention. This technique applies a simple yet vigorous approximation model to substitute expensive numerical simulation runs. However, almost all the existing methods independently use a single approximation model and neglect the potential synergies between these models. In order to make full use of the potential synergies of these existing approximation models, a novel multi-fidelity (MF) surrogate-assisted multi-objective production optimization (MOPO) method based on transfer stacking (MFTS-MOPO) is proposed. In the MFTS-MOPO method, the radial basis function network and support vector regression surrogate models are applied to approximate the high-fidelity (HF) model as the two additional low-fidelity (LF) models. Then a multi-fidelity surrogate model is adopted to evaluate objectives during the optimization process by transferring the two additional and streamline low-fidelity models to the computationally expensive high-fidelity model. Furthermore, two sampling infill strategies are applied to efficiently improve the quality of the multi-fidelity surrogate model. The uniqueness of the proposed MFTS-MOPO method is that the transfer stacking technique is employed to efficiently use the information from different fidelity models to establish the MF surrogate model and the infill sampling strategy used to improve its performance. In addition, three benchmark problems and two reservoirs with different scales were applied to illustrate the effectiveness and accuracy of the MFTS-MOPO method. It was found that the MFTS-MOPO method had superior performance in convergence and diversity than other conventional methods.