Multi-Objective Optimization of Composite Shielding for PGNAA Landmine Detection Based on NSGA-II

To address the conflict between detector-side background interference and shielding lightweight requirements in a PGNAA landmine detection system, a multi-objective optimization study was conducted for the layer thicknesses of a detector-side W-BPE-Pb composite shield. A Geant4 simulation model was established, and optimal Latin hypercube sampling was used to generate training samples. A PSO-BP neural-network surrogate model was then constructed to predict the normalized fast-neutron and gamma backgrounds. On this basis, the NSGA-II algorithm was introduced to optimize the two background responses and shield mass simultaneously. The results show that the surrogate model can describe the nonlinear relationship between shield thickness and detector-side background response. The Pareto solution set reveals the trade-off between background suppression and lightweight design. Characteristic-peak SNR validation indicates that different shielding schemes affect peak distinguishability differently, providing a reference for PGNAA shielding design.