Inverse Uncertainty Quantification for Subchannel Code With PSBT Experimental Benchmark

Abstract Inverse uncertainty quantification analysis for subchannel programs has been conducted to decide uncertainty distribution of important model parameters based on the modified Markov Chain Monte Carlo (MCMC) algorithm using the PSBT void fraction experiment. The posterior probability distribution expression for input parameter uncertainty can be derived using Bayesian principles, typically representing a proportional relationship. However, obtaining the direct normalization coefficient proves challenging, prompting the use of the MCMC algorithm to address this issue. Given the substantial external forward program calculations associated with stochastic methods, the Back-propagation Neural Network (BPNN) is employed to construct a surrogate model, serving as a substitute for intricate forward calculations. To significantly enhance the precision of the surrogate model, an adaptive approach based on relative entropy minimization is implemented to densify training sample points. The uncertainty of key input model parameters (slip ratio and turbulent mixing coefficient) was quantified, revealing that the uncertainty of the slip ratio is the primary source of void fraction uncertainty. After obtaining information on input parameter uncertainty, forward uncertainty analysis based on input uncertainty was conducted. The resulting uncertainty band for void fraction was obtained under a tolerance limit of 95%-95%. The results indicate that the uncertainty band effectively envelops the experimental data for void fraction. Using the statistical mean of parameter uncertainty obtained, the baseline values were calibrated. The predicted results of the model correction values were found to be more accurate than those of the baseline predictions.