A neutron transport calculation method based on Monte Carlo–Deep learning–interface current coupling and its validation in assembly problems

Neutron transport calculation is fundamental to reactor physics analysis and reactor design. Monte Carlo methods offer high accuracy and strong adaptability to complex geometries and energy spectra, but they usually involve high computational costs in pin-resolved reactor core calculations. In this study, a neutron transport calculation method based on Monte Carlo–Deep Learning–Interface Current (MC-DL-IC) coupling is proposed. In this method, composable pin-cell basic units serve as local response objects. Monte Carlo simulations generate local response data under incident partial-current conditions at the interfaces, neural networks are trained to learn the incident-to-outgoing response relationship of pin-cell basic units, and global coupling among multiple pin-cell basic units is achieved via the continuity condition of interface partial currents. The proposed method is tested on a one-group macroscopic cross-section assembly problem and an assembly problem with real materials. Numerical results show that the deviation in the effective multiplication factor is below 200 pcm, and the relative deviations in the flux and fission source distributions do not exceed ±1%. These results demonstrate the proof-of-concept validity and potential applicability of the proposed method.