Data-driven prediction and analysis method for nanoparticle transport behavior in porous media

Abstract

Engineering nanoparticles, as one of the application tools of nanotechnology, their transport behavior is closely related to applications such as reservoir sensing and environmental protection. Therefore, it is necessary to develop a general method to predict and analyze the nanoparticle transport behavior. In this paper, a data-driven prediction and analysis method for nanoparticle transport behavior in porous media is proposed. Firstly, a dataset of nanoparticle transport containing 411 samples is established, in which the missing data are effectively filled by random forest combining one-hot encoding. Then, a categorical boosting algorithm combined with synthetic minority oversampling technique is used to predict the retention fraction and retention profile. Finally, the Shapley additive explanation (SHAP) method is adopted to analyze feature significance. The results show that the proposed method has good performance on the prediction of nanoparticle transport behavior which is described by retention fraction and retention profile. At the same time, the interpretability of the SHAP method in analyzing nanoparticles transport behavior is also verified, which provides a new perspective for the further research and application.