The failure probability calculation of the passive system EPS in HPR1000 under LBLOCA based on the importance sampling algorithm

Passive safety systems play a critical role in reactor accident mitigation by relying on inherent physical mechanisms rather than active components. Although passive systems are generally considered more reliable than conventional active safety systems, their failure mechanisms involve significant uncertainties, making accurate reliability quantification challenging for traditional probabilistic safety assessment (PSA) methods. In particular, under severe accident conditions such as large break loss-of-coolant accidents (LBLOCA), the failure probability of passive systems is extremely low, while strong phenomenological uncertainties further complicate reliable estimation. To address these challenges, this study applies an importance sampling (IS) – based method to calculate the failure probability of the Emergency Passive System (EPS) in the HPR1000 reactor. The approach leverages established IS techniques and adapts them to the EPS context, improving both computational efficiency and estimation accuracy, and enabling the quantification of extremely small failure probabilities with a limited number of simulation samples. Furthermore, three methods – direct importance sampling, importance sampling combined with artificial neural networks (IS-ANN), and subset simulation combined with neural networks (SS-ANN) – are systematically compared. The results demonstrate that the failure probabilities obtained using the proposed IS-based method are in good agreement with those derived from neural network–enhanced approaches. This confirms the effectiveness and superiority of the proposed method for evaluating the reliability of passive safety systems and provides a practical and efficient framework for reliability assessment under severe accident conditions.