Abstract: In order to optimize the design of neutron shielding materials, a study is carried out with a homogeneous mixture of composite materials consisting of boron carbide, iron, tungsten, lead, bismuth oxide, polyethylene and NBS concrete. The evolutionary multi-objective optimization design of material composition and thickness is performed. Based on the simulation results of MCNP5, an adaptive RBF neural network dose prediction model is trained. The reference point-based non-dominated sorting genetic algorithm NSGA-Ⅲ is used to optimize three objective functions: weight, volume, and shielding property of the shielding material. The Pareto-optimal solution set is analyzed to verify the feasibility of the optimization method and provide methods and theoretical guidance for the multi-objective optimization design of neutron shielding materials.

Key words: multi-objective optimization, neutron shielding, Monte Carlo, RBF neural network