Radial Basis Function Surrogate Model-Based Optimization of Small Light-Weight Lead-Bismuth Reactor Core

Abstract

Owing to their high miniaturization potential and strong natural circulation capability, lead-bismuth reactors have been widely employed for nuclear energy utilization as well as military and civilian applications. However, there is a need to optimize the design of these reactors. To solve the complex high-dimensional nonlinear single-/multiobjective optimization problem of lead-bismuth reactors, we proposed a single-/multiobjective optimization method (termed as RBF-OLHS-NGA/NSGA II method) for optimizing the reactor core, which is based on the radial basis function (RBF) surrogate model (for prediction), orthogonal Latin hypercube sampling, niche genetic algorithm (for single-objective optimization), and nondominated sorting genetic algorithm (for multiobjective optimization). In view of this approach, the design optimization procedure of the lead-bismuth reactor based on RBF (DOPPLER-R) was developed, which combined the reactor Monte Carlo (RMC) code and the steady-state thermo-hydraulic analysis calculation (STAC) code to sample, predict, and optimize reactor core parameters. Furthermore, the single-/multiobjective optimization approach was verified by considering the core fuel loading and active zone volume of SPALLER-4 as the optimization objectives. The results show that the RBF surrogate model can accurately and rapidly predict the core characteristic parameters of lead-bismuth reactors. When compared to the value calculated using the RMC code, the relative error associated with the predicted effective multiplication factor ( $k_{eff}$ ) is within ±0.1%. Compared to the unoptimized values, the single-objective optimized fuel loading reduced by 400 kg, multiobjective optimized fuel loading decreased by 455–493 kg (optimization ratio of 78%–84%), and active zone volume reduced by 166362–182888 cm3 (optimization ratio of 72%–79%). These results prove that the proposed optimization method is feasible, exhibits high efficacy, and can provide new technical ideas for single-/multiobjective optimization of multiphysics, multivariable, and multiconstraint lead-bismuth reactors.