Predicting distribution coefficient and effective diffusion coefficient of radionuclides in bentonite: Multi-output neural network simulation and diffusion experimental study.
Journal:
Journal of hazardous materials
Published Date:
Mar 5, 2025
Abstract
Distribution coefficient (K) and effective diffusion coefficient (D) are two critical parameters for predicting the radionuclides diffusion in high-level radioactive waste (HLW) repositories. A novel framework using multi-output learning with an Artificial Neutral Network (ANN) was presented to simultaneously predict both parameters with a single model. A Generative Adversarial Network (GAN) algorithm was employed to augment data by producing pseudo-instances. The predictive accuracy of GAN-ANN model reached the determine coefficient (R) of 0.98 for K and 0.97 for D using a dataset of 26 input features and 2068 instances, including both experimental (1034) and pseudo-data (1034). Shapley Additive Explanations (SHAP) analysis identified the total porosity as the primary predictors for K and D, respectively. To assess the model's generalization capability, through-diffusion experiments were conducted on ReO, CoEDTA, and HSeO in compacted Gaomiaozi (GMZ) bentonite and illite/smectite mixed layers (I/S) at compaction densities ranging from 1200 to 1800 kg/m³ . The GAN-ANN model demonstrated robust performance in predicting radionuclide sorption and diffusion, with experimental-to-predicted K and D ratios ranging from 0.7 to 1.1 and from 0.7 to 2.9, respectively. The study offers a valuable predictive tool and a robust dataset to facilitate the safety assessment of HLW repositories.
Authors
Keywords
No keywords available for this article.
