A data-physical dual-driven surrogate model for reservoir simulation

The construction of seepage surrogate models is the frontier of simulation technology research for oil and gas reservoirs. However, the currently widely used pure data-driven seepage surrogate models have no theoretical support and require high data volume and data quality, which greatly limits the development of seepage surrogate models. Therefore, a dual-driven seepage proxy model integrating data-driven and physical-driven is proposed. Based on the pure data-driven seepage surrogate model, it integrates the seepage theory to simulate and predict the oil and gas seepage process. The results show that, compared with the pure data-driven model, even if the training data are extremely sparse, the dual-driven seepage surrogate model can still maintain high prediction accuracy. Second, the robustness of the dual-driven model is explored by adding different levels of noise interference to the training data, and it is verified that it is better than the pure data-driven seepage surrogate model. Finally, through transfer learning, the trained dual-driven seepage surrogate model is applied to the new seepage field, and the model can achieve rapid convergence and save computing resources.