Neutron Spectrum Unfolding using the Lasso Regression Method

Abstract

Neutron spectrum unfolding is used to adjust a known distribution in a reactor that is known to contain uncertainty. This known neutron spectrum is commonly determined in high resolution using transport codes such as MCNP. Measured reaction probabilities can be obtained through neutron activation analysis and can be used to adjust the simulated neutron spectrum. A linear system is defined that represents a relationship between the observed reaction probabilities and the calculated reaction probabilities that are a result of the simulated spectrum, and can be solved for an adjustment. The purpose of this investigation is to create a linear model with a simulated neutron spectrum and measured reaction probabilities, and to solve the model using the Lasso regression in the form of an elastic net regression. It uses benchmarks provided by the IAEA REAL-201X program to adjust the spectrum of the Annual Core Research Reactor in varying configurations to complete a proof of concept.