Transformation for Physical Interpretability and Data Refinement in a Machine-Learned NLTE Model of ICF Simulation*

The integration of machine learning techniques in Inertial Confinement Fusion (ICF) simulations has emerged as a powerful approach for enhancing computational efficiency. For example, by replacing the costly Non-Local Thermodynamic Equilibrium (NLTE) model with machine learning models, significant reductions in calculation time have been achieved [1, 2]. However, ensuring a comprehensive understanding of the physical implications and meaningfulness of the training data is crucial in this context. In this presentation, we propose two parameters, namely net emission (δ) and modified mean free path (τ), which are transformed from radiative opacity and emissivity, to address these challenges. Through these parameters, we explore the correlation between machine learning errors and physical quantities and assess the significance of opacity under simulation conditions. Our findings shed light on the physical impact of machine learning models on ICF simulations and offer insights into refining training data for improved efficiency.