Predicting anion diffusion in bentonite using hybrid machine learning model and correlation of physical quantities.

Area of Science:

• Geochemistry and Materials Science
• Environmental Science and Engineering
• Computational Science

Background:

• Radionuclide diffusion in bentonite is crucial for nuclear waste disposal safety.
• Understanding the link between bentonite's microstructure and diffusion is key to accurate modeling.
• Existing models often lack comprehensive integration of micro- and mesoscopic features.

Purpose of the Study:

• To develop and validate a machine learning model predicting radionuclide diffusion coefficients in bentonite.
• To elucidate the correlation between bentonite's micro/mesoscopic structure and radionuclide diffusion.
• To identify key factors influencing radionuclide diffusion using advanced analytical techniques.

Main Methods:

• Utilized a Light Gradient Boosting Machine (LightGBM) for predictive modeling.
• Optimized LightGBM hyperparameters with the Particle Swarm Optimization (PSO) algorithm.
• Incorporated micro- (ionic radius, montmorillonite stacking) and mesoscopic (porosity, conductivity) features.
• Validated predictions with through-diffusion experiments for HCrO4-, I-, and CoEDTA2-.

Main Results:

• The PSO-LightGBM model accurately predicted effective diffusion coefficients.
• Compacted dry density, ionic diffusion coefficient in water, ionic radius, and total porosity were identified as primary influencing factors.
• Shapley additive explanation and partial dependence plots provided insights into factor influence and relationships.
• Experimental validation confirmed the model's reliability and predictive accuracy.

Conclusions:

• Machine learning, integrating multi-scale features, enhances understanding of radionuclide diffusion mechanisms in bentonite.
• The developed model effectively links bentonite's microstructure to radionuclide transport behavior.
• This approach offers a reliable tool for assessing contaminant migration in geological formations.