Uncertainty Quantification on Spent Nuclear Fuel with LMC

TL;DR

This paper applies the Lasso Monte Carlo (LMC) method to quantify uncertainties in spent nuclear fuel calculations, demonstrating it is unbiased, more accurate, and computationally efficient compared to traditional methods.

Contribution

The paper introduces the application of LMC to nuclear fuel uncertainty quantification, showing its advantages in high-dimensional, computationally intensive scenarios.

Findings

LMC provides unbiased uncertainty estimates.

LMC achieves higher accuracy than standard Monte Carlo.

LMC reduces computational costs significantly.

Abstract

The recently developed method Lasso Monte Carlo (LMC) for uncertainty quantification is applied to the characterisation of spent nuclear fuel. The propagation of nuclear data uncertainties to the output of calculations is an often required procedure in nuclear computations. Commonly used methods such as Monte Carlo, linear error propagation, or surrogate modelling suffer from being computationally intensive, biased, or ill-suited for high-dimensional settings such as in the case of nuclear data. The LMC method combines multilevel Monte Carlo and machine learning to compute unbiased estimates of the uncertainty, at a lower computational cost than Monte Carlo, even in high-dimensional cases. Here LMC is applied to the calculations of decay heat, nuclide concentrations, and criticality of spent nuclear fuel placed in disposal canisters. The uncertainty quantification in this case is crucial to reduce the risks and costs of disposal of spent nuclear fuel. The results show that LMC is unbiased and has a higher accuracy than simple Monte Carlo.