Enhancing radiological risk evaluation through AI and HotSpot code integration: A Comparative study of LOCA and SGTR

Abstract

This study assesses and compares the radiological risks posed by two nuclear accident scenarios: Loss of Coolant Accident (LOCA) and Steam Generator Tube Rupture (SGTR). Using consistent environmental parameters and atmospheric conditions with radionuclide concentration levels specific to each scenario, the study evaluates radionuclide deposition rates and the Total Effective Dose Equivalent (TEDE). Key factors such as soil surface roughness and distance from the incident site are considered, as they significantly influence radionuclide deposition and dose rates. For accident management within the Nuclear Power Plant (NPP), an Artificial Neural Network (ANN) model demonstrated 95.51% accuracy in fault prediction, enhancing operational reliability in emergencies. Integrating Artificial Intelligence (AI), specifically the Long Short-Term Memory (LSTM) model, with the HotSpot Health Physics Code enabled enhanced prediction of ground surface deposition dose (GSD) at distances beyond traditional limits. At a distance of 200 km, HotSpot’s dose calculation recorded GSD values of 1.6 × 10⁻³ kBq m−2 for LOCA, and 9.5 × 10⁻⁴ kBq m−2 for SGTR, respectively, the LSTM model forecasted significantly lower GSD values at greater distances, reaching 1.3 × 10⁻⁶ kBq m⁻2 at 270 km for LOCA and 5.58 × 10⁻⁸ kBq m⁻2 at 220 km for SGTR. Results show that radionuclide deposition decreases with increased soil roughness, with LOCA scenarios generally yielding higher TEDE values across zones compared to SGTR. At a soil roughness level of 3 cm, LOCA deposition reached 0.25 kBq m⁻2 in outer zones, compared to 0.11 kBq m⁻2 for SGTR. These findings underscore the importance of tailored planning strategy protocols based on terrain and proximity factors for effective incident management.