Analyses of the bias and uncertainty of SNF decay heat calculations using Polaris and ORIGEN

Abstract

The bias and uncertainty of calculated decay heat from spent nuclear fuel (SNF) are essential for code validation. Also, predicting these quantities is crucial for deriving decay heat safety margins, influencing the design and safety of facilities at the back end of the nuclear fuel cycle. This paper aims to analyze the calculated spent nuclear fuel decay heat biases, uncertainties, and correlations. The calculations are based on the Polaris and ORIGEN codes of the SCALE code system. Stochastically propagated uncertainties of inputs and nuclear data into calculated decay heats are compared. Uncertainty propagation using the former code is straightforward. In contrast, the counterpart of ORIGEN necessitated the pre-generation of perturbed nuclear cross-section libraries using TRITON, followed by coincident perturbations in the ORIGEN calculations. The decay heat uncertainties and correlations have shown that the observed validation biases are insignificant for both Polaris and ORIGEN. Also, similarities are noted between the calculated decay heat uncertainties and correlations of both codes. The fuel assembly burnup and cooling time significantly influence uncertainties and correlations, equivalently expressed in both Polaris and ORIGEN models. The analyzed decay heat data are highly correlated, particularly the fuel assemblies having either similar burnup or similar cooling time. The correlations were used in predicting the validation bias using machine learning models (ML). The predictive performance was analyzed for machine learning models weighting highly correlated benchmarks. The application of random forest models has resulted in promising variance reductions and predicted biases significantly similar to the validation ones. The machine learning results were verified using the MOCABA algorithm (a general Monte Carlo-Bayes procedure). The bias predictive performance of the Bayesian approach is examined on the same validation data. The study highlights the potential of neighborhood-based models, using correlations, in predicting the bias of spent nuclear fuel decay heat calculations and identifying influential and highly similar benchmarks.