S_N-MscaleDNN: A coupling approach for rapid shielding-scheme evaluation of micro gas-cooled reactor in the large design-parameter space

Rapid shielding-scheme evaluation in the large shielding-design parameter space (LSDPS) helps to optimize much better reactor shielding schemes. For this reason, a practical engineering approach utilizing a 1-D S_N calculation and MscaleDNN network model is presented for accelerating reactor shielding-scheme evaluation in the LSDPS. We first review two commonly-used neural network-based methods for shielding-scheme evaluation and analyze their difficulties in the LSDPS through theoretical machine learning and data transformation analysis. Based on these discussions, the work transitions from method reviews to our two-step coupling solution to overcome these difficulties arising from high-frequency and multi-scale characteristics in the task. We first adopt equivalent 1-D S_N calculations to quickly obtain inaccurate radiation dose rates at target points and then correct them by the frequency-scaled neural network MscaleDNN. The numerical results demonstrate that the proposed method achieves engineering-acceptable evaluation accuracy and higher accuracy than the conventional methods in the LSDPS.