Enhancement of random sampling by a combined approach of control variates and Latin hypercube sampling for uncertainty quantification in light water reactor lattice calculations

Fujita, Tatsuya  

This study confirmed the efficiency of a combined approach of the control variates (CV) and the Latin hypercube sampling (LHS), which enhanced the random-sampling-based uncertainty quantification due to cross-section (XS) covariance data, by considering the effect of statistical variation and also performed the sensitivity analyses on the influence due to the selection of alternative parameter to apply CV. The convergence performance for the uncertainty of infinite multiplication factor (k-infinity) during the random sampling was compared between several efficient sampling techniques such as the antithetic sampling (AS), LHS, CV, and the combined approaches of them in the PWR-UO fuel assembly geometry. The k-infinity uncertainty was evaluated by statistically processing several times Serpent2 calculations using perturbed ACE-formatted XS files based on ENDF/B-VIII.0. CV+LHS was more efficient than AS, LHS, and CV+AS. In addition, sensitivity analyses were performed to select alternative parameters used in CV. The 33 mini fuel lattice calculation can improve the efficiency of CV+LHS. The reason was qualitatively considered that this calculation can capture the influence of XS covariance data for Gd isotopes. Consequently, the applicability of CV+LHS for the improvement of convergence performance to evaluate the k-infinity uncertainty during the random sampling was confirmed.