A unified intelligent model for estimating the (gas plus n-alkane) interfacial tension based on the eXtreme gradient boosting (XGBoost) trees
A unified intelligent model for estimating the (gas plus n-alkane) interfacial tension based on the eXtreme gradient boosting (XGBoost) trees
复制标题
基于极限梯度提升 (XGBoost) 树的统一智能模型，用于估计（气体加正构烷烃）界面张力
DOI：
10.1016/j.fuel.2020.118783
复制
发表时间：
2020
期刊：
影响因子：
7.4
通讯作者：
Wu Kuankuan
中科院分区：
文献类型：
--
作者：
Zhang Jiyuan;Sun Yanchun;Shang Lin;Feng Qihong;Gong Lirong;Wu Kuankuan
注入气与主体油之间的界面张力(IFT)是影响最终驱油效率和注气驱油效果的关键参数。准确表征不同正构烷烃与注入气体之间的界面张力，对于更深入地了解(天然气+石油)体系界面行为背后的主导机制，进而确保注气提高采收率项目的优化设计至关重要。IFT的实验室测量通常需要昂贵的实验仪器、耗时的操作程序和繁琐的数据推导。本文提出了一种新的监督学习(SL)方法，即极值梯度增强(XGBoost)树，用于(气体+正构烷烃)IFT的快速估计。在一个由1561个数据集组成的大型数据库的基础上，建立了不同注入气体种类和正构烷烃的统一估计模型。结果表明，统一模型能够准确地再现基于压力、温度、正构烷烃相对分子质量和气体组成的实验IFT。在精度和稳健性方面，新模型优于多层感知器(MLP)、支持向量回归(SVR)和已有的相关性。此外，应用排列重要度(PI)来量化每个输入特征对IFT的重要性，得出特征对(气体+正构烷烃)IFT重要性递减的排序为：压力≫、正构烷烃相对分子质量、气体组成和温度。
The interfacial tension (IFT) between the injecting gas and host oil is a key parameter that affects the ultimate displacement efficiency and gas injection enhanced oil recovery (EOR) performance. The accurate characterization of the IFT between varying n-alkanes and injecting gases is crucial to obtaining deeper insight into the predominating mechanisms behind the interfacial behaviors of (gas + oil) systems, and in turn ensuring the optimal design of gas injection EOR projects. Laboratory measurement of the IFT usually requires expensive experimental apparatus, time-consuming operation procedure and cumbersome data deduction. This paper proposed the use of a novel supervised learning (SL) method, namely the eXtreme gradient boosting (XGBoost) trees, for the fast estimation of (gas + n-alkane) IFT. A unified estimation model was constructed for varying injecting gas species and n-alkanes based on a large database consisting of a number of 1561 data sets. Results showed that the unified model is capable of accurately reproducing the experimental IFT based on pressure, temperature, n-alkane molecular weight and gas composition. It was also demonstrated that the new model outperforms the multi-layer perceptron (MLP), support vector regression (SVR) and existing correlations in terms of accuracy and robustness. Furthermore, the permutation importance (PI) was applied to quantify the importance of each input feature to the IFT, which concluded that the ranking of features in terms of decreasing importance to the (gas + n-alkane) IFT are: pressure ≫ n-alkane molecular weight > gas composition > temperature.