Simulation-Based Artificial Intelligence Framework for Assessing Environmental Impacts of Radioactive Releases from Nuclear Facilities: A Real-World Case Study

This research presents a novel framework leveraging artificial intelligence for the accurate estimation of radiation doses resulting from the release of radioactive materials at nuclear facilities. The methodology couples HOTSPOT with an Adaptive Neuro-Fuzzy Inference System (ANFIS) developed in MATLAB, enabling rapid and sophisticated predictions of population exposure doses arising from atmospheric radionuclide dispersion. The framework is validated through a case study of hypothetical failure scenarios of the radioactive waste (radwaste) system at the Bushehr Nuclear Power Plant (BNPP), Iran, including: (I) equipment leakage or malfunction in the gaseous radwaste system, (II) leakage or rupture of liquid radwaste storage tanks, and (III) simultaneous failure of both gaseous and liquid subsystems, analyzing the spatial distribution of radionuclide concentrations at various distances from the emission source. Dose estimations for critical human organs, including the whole-body, thyroid, and lungs, are derived using both HOTSPOT simulations and the trained ANFIS model. Comparative analysis demonstrates excellent agreement between ANFIS predictions and HOTSPOT results, with mean absolute percentage errors below 0.01 and coefficients of determination exceeding 0.98 across all scenarios, alongside a 97% reduction in computational time relative to full deterministic simulations. In addition, the final ANFIS–HOTSPOT model was evaluated and validated using real monitored data under normal operation conditions, confirming its predictive accuracy in routine facility operations. Results indicate that, under both hypothetical accident scenarios and normal operation, the maximum annual effective dose to the public remains well below regulatory limits (0.05 mSv/year), which is substantially lower than the regulatory thresholds established by the International Commission on Radiological Protection (ICRP) and the International Atomic Energy Agency (IAEA). Overall, the developed framework is highly scalable and adaptable, providing an efficient and reliable tool for modeling radionuclide transport and assessing radiological doses across all exposure pathways, supporting both operational safety and emergency preparedness.