APPROXIMATE ESTIMATION OF ITERATED FISSION PROBABILITY BY DEEP NEURAL NETWORK

Abstract

Iterated fission probability (IFP), which is proportional to a fundamental mode of adjoint angular neutron flux, has increasingly been used as a weighting function in Monte Carlo calculations. The Monte Carlo IFP methods stochastically estimate IFP for a given phase-space location. In this work, we investigated the applicability of a deep neural network for approximating an unknown underlying function, which maps from a phase-space location to an IFP in a given fissile system, from a dataset produced by a Monte Carlo IFP method. The preliminary application has been performed for the Godiva core and the comparison showed a varying degree of agreement and discrepancy between the estimated IFPs by the DNN and the reference adjoint angular neutron flux by a deterministic neutron transport code PARTISN.