An assembly-level neutronic calculation method based on LightGBM algorithm

Abstract

In this paper, a novel assembly-level neutronic calculation method based on a machine learning technique, the LightGBM algorithm, has been proposed. The calculating model based on LightGBM algorithm is established to calculate the assembly-level neutronic parameters, including Infinite Multiplication Factor ‘Kinf’, Power Peak Factor ‘PPF’, and Burnup ‘B’, which are normally calculated by exactly solving the neutron transport equations. The proposed method was first applied to assemblies cases of different sizes varying from 3 × 3, 5 × 5, 7 × 7, 9 × 9, 14 × 14 and 17 × 17 to calculate Kinf, and compared with other five Machine Learning algorithms. The results show that the proposed method based on LightGBM Algorithm has the most stable and decent performance in forecasting Infinite Multiplication Factor ‘Kinf’. Therefore, LightGBM was directly implemented to calculate other neutronic calculations for a 17 × 17 assembly. The overall precision is also satisfactory. The main distribution range of errors of Kinf is [−0.003, +0.0084], of PPF is [−0.0009, +0.0153] and of Burnup (MWD/KgU) is [−0.0628, +0.2552]. Overall, the new proposed method will be promising in some certain application scenarios especially during the Loading Pattern (LP) optimization.