Abstract This study aims to develop a grid-based smart proxy model (G-SPM) for water flooding enhancement under different injection/production scenarios. The contribution of this work is combining experimental design and data mining techniques to prepare a high-quality database and introducing a hyper-feature to incorporate flow physics into database. The proxy was constructed by training a sequential neural network model. Validating G-SPM on three blind cases showed that oil saturation in grid blocks is accurately predicted in shorter time compared to numerical simulator. This study also revealed that data from one previous time step is enough to be used in G-SPM training.

中文翻译：

---

开发基于网格的智能代理模型来评估各种注水场景

摘要 本研究旨在开发一种基于网格的智能代理模型（G-SPM），用于不同注入/生产情景下的注水增强。这项工作的贡献是结合实验设计和数据挖掘技术来准备一个高质量的数据库，并引入一个超特征将流物理纳入数据库。代理是通过训练序列神经网络模型构建的。在三个盲案例上验证 G-SPM 表明，与数值模拟器相比，可以在更短的时间内准确预测网格块中的含油饱和度。这项研究还表明，前一个时间步的数据足以用于 G-SPM 训练。