Radiation Shielding Optimization Design Based on PK-MC Integrated Multi-Fidelity Surrogate Model

While the low-fidelity (LF) method like the Point-Kernel Integration (PK) method, though fast, lacks sufficient accuracy to guarantee reliable designs.

To address this accuracy-efficiency trade-off, this paper proposes a PK-MC Integrated Multi-fidelity surrogate model for radiation shielding optimization, aiming to achieve high-precision optimization at lower computational costs. In this paper, to eliminate fidelity errors caused by mismatched sample spaces, this study adopts an Optimal Latin hypercube sampling(OLHS) strategy with inheritance, ensuring that HF and LF samples share a unified parameter space; and it introduces logarithmic scaling preprocessing, enhancing the model’s ability to capture the data features of radiation exponential decay; Finally, it constructs a Co-kriging framework that integrates a large amount of LF data from the PK method and a small amount of HF data from the MC method, thereby leveraging efficiency of PK and precision of MC for high-performance modeling.