Developing a machine learning-based predictive model for cesium sorption distribution coefficient on crushed granite.

Journal: Journal of environmental radioactivity
PMID:

Abstract

The sorption of radionuclides on granite has been extensively studied over the past few decades due to its significance in the safety assessment of geological disposal for high-level radioactive waste (HLW). The sorption properties of granite for radionuclides exhibit considerable variability under different experimental conditions. To reduce the time and cost associated with traditional experiments, this study developed a data-driven approach utilizing machine learning (ML) algorithms to predict the sorption distribution coefficients of cesium (Cs) on crushed granite efficiently. Four ML algorithms, namely AdaBoost, GBDT, LightGBM, and XGBoost, were employed to construct predictive models using a dataset of 384 data points. All models demonstrated strong performance, with R values exceeding 0.8 for both the training and test sets. Comparative analysis of evaluation metrics indicated that the XGBoost model exhibited the best predictive performance and generalization ability. An explanation analysis of the XGBoost model further revealed the importance and influence of each input feature in predicting the distribution coefficient of Cs on crushed granite. The features affecting radionuclide sorption on granite were ranked by importance as follows: solid/liquid ratio, ion strength, pH, contact time, initial concentration, and maximum particle size. The underlying sorption mechanisms by which different input features affect the sorption coefficient, as derived from shapley additive explanations (SHAP) analysis, correspond with experimental observations. The approach proposed in this study can serve as a supplement to resource-intensive experimental methods, providing new insights into predicting the sorption behavior of radionuclides on crushed granite for the safety assessment of HLW geological disposal.

Authors

  • Funing Ma
    School of Environmental and Municipal Engineering, Qingdao University of Technology, Qingdao, 266520, China.
  • Zhenxue Dai
    College of Construction Engineering, Jilin University, Changchun 130026, China.
  • Fangfei Cai
    School of Architecture and Engineering, Qingdao Binhai University, Qingdao, 266555, China. Electronic address: caiff21@mails.jlu.edu.cn.
  • Xiaoying Zhang
    College of Veterinary Medicine, Northwest A&F UniversityYangling, China; Chinese-German Joint Laboratory for Natural Product Research, Qinling-Bashan Mountains Bioresources Comprehensive Development C.I.C., College of Biological Science and Engineering, Shaanxi University of TechnologyHanzhong, China.
  • Yue Ma
    The School of Civil Engineering, Harbin University, Harbin 150086, China.
  • Dayong Wang
    BeckLab, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts.