Overview of CNSC's Experience with Uncertainty Quantification in the Severe Accident Domain - Areas of improvement using artificial intelligence

Abstract of the journal article presented in:
Consultancy Meeting on the Use of Artificial Intelligence in Assessing the Uncertainties in Severe Accident Modelling and Simulation for Advanced Water-Cooled Reactors, Based on CRPI31033 Findings
March 21-23, 2023

Prepared by:
Mounia Berdaï
Canadian Nuclear Safety Commission

Abstract:

Uncertainty Quantification (UQ) in Severe Accidents (SA) aims to increase our confidence in severe accident codes’ predictions and enhance our responses to nuclear or radiological emergencies. For this purpose, Python Scripts for Uncertainty Quantification of MAAP-CANDU and MELCOR (PSUQM2) toolkit was developed to allow the coupling with MAAP-CANDU and MELCOR (severe accident codes) and performs uncertainty quantification for selected severe accident Figure Of Merit (FOM).

PSUQM2 can be used for a wide range of applications, such as assessing uncertainties around some unknown phenomena in Small Modular Reactors (SMR)s of all technologies and evaluate their safety margins.

Within the scope of PSUQM2 two new concepts were proposed: the Dynamic Probability Density Function (DPDF) and the Dynamic correlation between Uncertain Parameters (UPs) and FOMs. We also demonstrated that Boxplot could be a powerful tool that provide information on the characteristics of the distribution for a given FOM.

The main challenge encountered during the UQ was the characterization of the UPs with the appropriate probability distribution that is consistent with the physical nature of each UP. Moreover, the identification of the correlated UPs was not an easy task. Certain UPs that were assumed to be “uncorrelated”, UQ outcomes revealed the opposite.

To overcome some of these challenges, the use of Artificial Intelligence (AI) could be an effective alternative to predict the behavior of nuclear power plants for un-simulated scenarios. AI use could therefore be beneficial for the resolution of the gaps/challenges raised during the UQ exercise and to compensate for the lack of knowledge on certain phenomena in the SA domain.

To obtain a copy of the abstract’s document, please contact us at cnsc.info.ccsn@cnsc-ccsn.gc.ca  or call 613-995-5894 or 1-800-668-5284 (in Canada). When contacting us, please provide the title and date of the abstract.