Researchers from the Hefei Institutes of Physical Science of the Chinese Academy of Sciences have developed a neural network model based on self-attention mechanisms to rapidly predict radiation shielding designs for space reactors. 

The breakthrough, aimed at optimizing shielding configurations more efficiently, was recently published in Nuclear Engineering and Design.

Micro and small reactors are emerging as compact, safe, and low-carbon energy solutions, particularly for space missions. However, designing effective radiation shielding for these reactors poses significant challenges due to tight spatial constraints, strict weight limits, and complex material interactions. While traditional Monte Carlo simulations offer high accuracy, they are computationally intensive and time-consuming—making them less ideal for quick design iterations.

To address this bottleneck, the researchers focused on space reactors and developed an intelligent model to help design radiation shielding more quickly and efficiently. This model is based on "self-attention neural network," which can learn patterns and make accurate predictions. The model was trained using datasets generated by SuperMC, a sophisticated simulation tool developed by the institute that calculates radiation interactions with shielding materials.

Once trained, the model can rapidly evaluate input parameters such as shielding weight and radiation dose levels to propose optimized shielding configurations. Tests showed that the model's predictions deviated less than 3% from those of conventional Monte Carlo methods, but required significantly less computation time.

This study provides an innovative approach to shielding design optimization for micro and small reactors.

Neural network model diagram(Image by CHEN Qisheng)