Comparative analysis of different plasma models and machine learning approaches for accurate plasma-parameter estimation

The accurate estimation of plasma parameters is pivotal for comprehending the ionization balance and radiation dynamics in magnetic fusion and astrophysical plasmas. In the present work, we perform systematic calculations of the plasma parameters, such as average charge state, effective charge state, and radiative power loss, for helium, neon, and argon plasma species using the ATOMIC collisional-radiative (CR) code from the Los Alamos suite [1]. These parameters are calculated for a wide range of electron densities and temperatures covering coronal and local thermodynamic limits. A fine-structure resolved CR model is used for helium, while configuration-average CR models are implemented for neon and argon plasma species. The calculated values are then compared with the other plasma models, e.g., the superconfiguration-based CR model (FLYCHK [2]) and commonly used coronal equilibrium model, to demonstrate the advantages and limitations of each model. A crucial aspect of our work involves utilizing the calculated values to develop different machine learning models, namely, Random Forest Regression, XGBoost, and Deep Neural Network. These models are aimed to facilitate the rapid and accurate prediction of plasma parameters. The accuracy of the models is evaluated using metrics like mean-squared error and R² values. The results and tools obtained in this work offer valuable insights and a robust framework for estimating plasma parameters, contributing significantly to the ongoing advancements in magnetic fusion and astrophysical plasma research.

[1] C. J. Fontes et al., Journal of Physics B 48, 144014 (2015)

[2] H.-K. Chung et al., High Energy Density Physics 1, 3 (2005)