A Comparative study of efficient sampling techniques for uncertainty quantification due to cross-section covariance data

Fujita, Tatsuya  

The convergence process of the k-infinity uncertainty during random-sampling-based uncertainty quantification was compared between several efficient sampling techniques. The k-infinity uncertainty was evaluated by statistically processing several times of SERPENT 2.2.1 calculations using perturbed ACE files based on JENDL-5 cross-section covariance data. The antithetic sampling (AS), the Latin hypercube sampling (LHS), the control variates (CV), and the combination approaches of them were focused on in the present paper. In PWR-UO fuel assembly geometry without the nuclide depletion, as discussed in past studies, AS and LHS showed higher efficient convergence than nominal sampling without any efficient sampling techniques. In terms of CV, though a stand-alone application did not have a large impact on the k-infinity uncertainty convergence, its performance was improved in combination with AS, as discussed in the past study. In addition, a new combined approach of LHS and CV (CV+LHS) was proposed in the present paper. CV+LHS improved the k-infinity uncertainty convergence and was more efficient than CV+AS. The main reason for this improvement was that the convergence for the mean value of alternative parameters in CV was enhanced by applying LHS. Consequently, this study proposed the new combined approach of CV+LHS and confirmed its efficiency performance for the random-sampling-based uncertainty quantification in the PWR-UO fuel assembly geometry. The applicability of CV+LHS for the nuclide-depletion calculations will be confirmed in future studies.