Long-term prediction of radionuclide migration in bentonite: diffusion experiments and combined GAN-DNN-PINN model
摘要
Long-term diffusion of H79SeO3−, 99TcO4−, 238UO22+, and 237NpO2+ in compacted bentonite was investigated using a combined GAN-DNN and PINN model. The apparent diffusion coefficient (Da) of HSeO3− and ReO4− was measured using a through-diffusion method to validate the generalization of GAN-DNN. Subsequently, the PINN model simulated the spatiotemporal evolution of radionuclide concentrations over 1000-year using the GAN-DNN-predicted Da as inputs. It reveals that strong sorption effectively restricted 238UO22+ and 237NpO2+ diffusion, whereas anionic exclusion facilitated the penetration of 66%–86% of 99TcO4− and H79SeO3⁻ to penetrate the 0.36 m engineered barrier of the Beishan repository in China.