A new sampling scheme combining maximum entropy and moment matching techniques for reactor physics uncertainty quantification

The sampling method is widely used for the quantification of reactor physics analysis uncertainty, while the traditional statistical sampling methods often require a huge number of samples to ensure the stability of uncertainty quantification results. For this reason, a new two-step sampling scheme based on the combination of information entropy theory and moment matching technique is proposed for the quantification of the reactor physics uncertainty. We construct a convex programming model to optimize initial samples by simultaneously maximizing information entropy and minimizing moment deviations. The optimized samples can not only characterize the statistics of the real distribution better, but also control the information entropy with a smaller sample size. Moreover, a quantitative criterion for sample selection induced by the maximum entropy and moment matching is proposed. Finally, a series of empirical studies are carried out to show the capability and efficiency of the introduced two-step sampling method.