A non-intrusive reduced order model for the characterisation of the spatial power distribution in large thermal reactors

Abstract

Large thermal reactors endowed with heavy reflectors are very sensitive to localised perturbations. To adequately characterise the spatial effects of these localised disturbances on the reactor power and, thus, to study the spatial stability of the core in a computationally efficient manner, without modifying the legacy codes that may be inefficient for this kind of parametric safety analysis, a non-intrusive surrogate model to approximate the power distributions is proposed in this paper. The meta-model is composed of two steps. First, the group constants, originally produced with the Monte Carlo method for the full-core diffusion calculations, are approximated via a polynomial chaos expansion regression model, with an acceptable accuracy. Then, a model combining Proper Orthogonal Decomposition and Radial Basis Functions is trained to replace the expensive full-core diffusion calculations. Due to the spatial dependence of the input perturbations, some ad hoc strategies are proposed and successfully applied in the reduced order model training phase. Finally, the bootstrap technique is employed to assess the quality of the meta-model approximations and to provide an estimation of the modelling error distribution.