Application of Heterogeneous Ensemble Learning for CO<sub>2</sub>–Brine Interfacial Tension Prediction: Implications for CO<sub>2</sub> Storage
Bin Shen, Shenglai Yang, Jiangtao Hu, Yumeng Gao, Hang Xu, Xinyuan Gao, Hao Chen
Abstract
Carbon capture, utilization, and storage (CCUS) is a green engineering technology to reduce CO 2 emissions and mitigate climate warming. It is crucial to accurately predict the CO 2 –brine interfacial tension (IFT) in order to evaluate the carbon storage capacity of saline aquifers. Traditional experimental methods are time-consuming and costly. The existing empirical correlation methods of IFT have been found to be inaccurate. Instead, machine learning (ML) methods have a superior ability to predict IFT. However, the existing machine learning methods lack an in-depth examination of the main factors influencing IFT, as well as the simultaneous improvement strategy of accuracy and time cost and further reliability verification. In this paper, we first propose a heterogeneous ensemble learning IFT prediction model based on XGBoost and LightGBM. The new model is simultaneously optimized in terms of both accuracy and time cost. Our proposed model has been proven to be the most accurate and time-efficient through several comparative studies. The variable trend analysis, the leverage method, and Shapley values (SV) are also used to investigate the effectiveness and interpretability of the model. The density differences parameter is used for the first time as an input parameter in the model which is found to be an appropriate parameter. A potential law between temperature and IFT can also be derived from the new model. Considering the optimization of input parameters, time, and accuracy simultaneously and estimating the CO 2 storage capacity of saline aquifers through IFT are the main values of this research.