Physics-based simulation for fluid flow in porous media is a computational technology to predict the temporal-spatial evolution of state variables (e.g. pressure) in porous media, and usually requires high computational expense due to its nonlinearity and the scale of the study domain. This letter describes a deep learning (DL) workflow to predict the pressure evolution as fluid flows in large-scale 3D heterogeneous porous media. In particular, we apply feature coarsening technique to extract the most representative information and perform the training and prediction of DL at the coarse scale, and further recover the resolution at the fine scale by 2D piecewise cubic interpolation. We validate the DL approach that is trained from physics-based simulation data to predict pressure field in a field-scale 3D geologic CO_2 storage reservoir. We evaluate the impact of feature coarsening on DL performance, and observe that the feature coarsening can not only decrease training time by >74% and reduce memory consumption by >75%, but also maintains temporal error <1.5%. Besides, the DL workflow provides predictive efficiency with ~1400 times speedup compared to physics-based simulation.

翻译：对多孔介质中流体流体进行物理模拟是一种计算技术,用来预测多孔介质中国家变量(例如压力)的时间空间演变,通常需要较高的计算费用,因为其非线性和研究领域的规模。本信描述了一个深度学习(DL)工作流程,用来预测在大型3D多孔多孔介质中流体流体流体的压力演变。特别是,我们采用特征粗化技术来提取最具代表性的信息,在粗糙的尺度上进行DL的培训和预测,并在微小的规模上通过2D片立方间插进一步恢复分辨率。此外,我们验证了从基于物理的模拟数据中培训的DL方法,以预测外地规模3D地质CO_2储存库的压力场。我们评估了地貌分解对DL性能的影响,并观察到特性粗化不仅能够减少培训时间74%,并将记忆消耗减少>75%,而且还保持时间误差<1.5 %。此外,DL工作流程提供了以~1400倍的速度模拟到物理模拟的预测效率。