A simulation-based framework for scaling factor modeling and GMM-based zonal stratification in activated decommissioning waste

This study presents a simulation-based framework for evaluating scaling factor (SF) models in activated decommissioning waste, emphasizing the importance of material- and activation-dependent heterogeneity. Neutron flux and activation data for the core barrel (CB), thermal shield (TS), and reactor vessel (RV) were generated using MCNP and ORIGEN-S to form a theoretical population. Unlike operational waste, CB and TS exhibited segmented SF distributions driven by neutron flux gradients, deviating from the conventional log-normal assumption. As activation levels increased, distribution nonlinearity became more pronounced. Polynomial regression effectively captured these effects, and regression models outperformed mean-based models in predictive accuracy. However, under realistic uncertainties, their advantage in confidence interval width was lost and prediction stability declined. Mean-based models, while robust, consistently underestimated difficult-to-measure nuclide inventories. To address this, a Gaussian Mixture Model–based zoning method was introduced to enhance sampling representativeness. This approach reduced underestimation and confidence intervals—even with very small sample sizes—and became increasingly effective with larger samples. The RV showed highly uniform behavior due to low activation, suggesting it may be treated as homogeneous waste. As SF confidence intervals are highly sensitive to sampling and analytical uncertainty, sample sizes should reflect these uncertainty sources to ensure reliable inventory estimation.