Uncertainty Quantification of a Genetic Algorithm for Neutron Energy Spectrum Adjustment

Abstract

GenSpec is software designed to use a genetic algorithm for neutron energy spectrum adjustment. Currently GenSpec can produce adjusted spectra, but the corresponding covariance matrix is not produced. The uncertainty quantification process implemented includes a parametric sensitivity analysis of the genetic algorithm modifiers for population, generations, gene-sites, polynomial order, and mutation rate. A random perturbation analysis was used to characterize the covariance of the genetic algorithm using multivariate normal random sampling of the characterized input data. The produced 640 by 640 covariance matrix has retained some characteristic features of the sampled covariance. The uncertainty found in the GenSpec program has minimized the covariance present in a calculated trial spectrum.