裂缝性致密油气藏注气提高采收率的数值模拟研究

致密油气藏的开发成本高昂且产量递减很快，采收率远低于常规油藏，因而提高采收率极为必要。气体粘度远低于水且气体分子尺寸小，所以注气是提高致密油藏采收率潜力最大的方法之一。目前，注气已在北美和国内油田得到了广泛应用。众所周知，准确描述相态变化是模拟注气过程的关键。在致密油藏中，纳米孔隙对闪蒸计算的约束效应不可忽略，这在已有的数值模拟中虽有考虑但处理过于简单，不能反映实际储层中的孔径分布对相变的影响。另一方面，注气需要实现混相，压力较高，所以有效应力的变化对裂缝网络和基质的孔隙度和渗透率影响较大。现有的应力敏感关联式对毛管力和相渗曲线的修正仅使用单一的渗透率和孔径值，过于粗略，且在模拟注气中的应用相对缺乏。本课题旨在解决这两个难题，并进一步耦合研究孔径分布下的纳米约束效应和动态变化的孔渗应力敏感性。此外，机器学习可大幅提高闪蒸计算速度，本课题将对比选择数种机器学习算法用以提高模拟效率。

Investment in developing tight reservoirs is very high compared to conventional reservoirs. Moreover, well production rate declines rapidly, resulting in much lower recovery rate than conventional reservoirs. Therefore, it is imperative to develop highly-efficient EOR technologies. Seeing that gas viscosity is much lower than water, and gas molecules are much smaller than chemical flooding agent molecules, gas injection is one of the most promising EOR methods for tight oil and gas reservoirs. At present, gas injection EOR has been widely applied in North America and some domestic oilfields. First and foremost, accurate description of phase behavior is crucial to simulate gas injection processes. In tight oil and gas reservoirs, the nanoconfinement effect on flash calculation cannot be neglected due to the vast amount of nanopores. Although nanoconfinement has been considered in existing reservoir simulations, the numerical treatment is oversimplified and cannot reflect the effect of pore size distribution on phase equilibrium in real reservoir rocks. Secondly, high operation pressure of gas injection is generally required to achieve miscibility, so the porosity and permeability of fracture network and tight reservoir matrix are greatly affected by effective stress changes due to gas injection. However, most existing relative permeability and capillary pressure correlations accounting for stress sensitivity only use single absolute permeability and pore size values to make corrections, which is inaccurate. And application of accurate correlations in simulating gas injection EOR is relatively rare. This project aims to solve these two problems, and further study the coupling effects of the nanoconfinement exerted by pore size distribution and the dynamic stress-dependent rock property. In addition, this project will compare different machine learning algorithms and select the optimal ones to improve the efficiency of gas injection simulation, given their demonstrated advantages in improving flash calculation in recent studies.