Uncertainty analysis of dynamic PRA using nested Monte Carlo simulations and multi-fidelity models

Zheng, X.; 玉置 等史; 高原 省五; 杉山 智之; 丸山 結

Proceedings of Probabilistic Safety Assessment and Management (PSAM16) (Internet), 10 Pages, 2022/09

https://www.iapsam.org/PSAM16/

Uncertainty gives rise to the risk. For nuclear power plants, probabilistic risk assessment (PRA) systematically concludes what people know to estimate the uncertainty in the form of, for example, risk triplet. Capable of developing a definite risk profile for decision-making under uncertainty, dynamic PRA widely applies explicit modeling techniques such as simulation to scenario generation as well as the estimation of likelihood/probability and consequences. When quantifying risk, however, epistemic uncertainties exist in both PRA and dynamic PRA, as a result of the lack of knowledge and model simplification. The paper aims to propose a practical approach for the treatment of uncertainty associated with dynamic PRA. The main idea is to perform the uncertainty analysis by using a two-stage nested Monte Carlo method, and to alleviate the computational burden of the nested Monte Carlo simulation, multi-fidelity models are introduced to the dynamic PRA. Multi-fidelity models include a mechanistic severe accident code MELCOR2.2 and machine learning models. A simplified station blackout (SBO) scenario was chosen as an example to show practicability of the proposed approach. As a result, while successfully calculating the probability of large early release, the analysis is also capable to provide uncertainty information in the form probability distributions. The approach can be expected to clarify questions such as how reliable are results of dynamic PRA.