KN1DPy and Supervised Machine Learning to Predict Neutral Density Profiles

Predicting neutral profiles is critical for understanding fusion reactor fueling requirements [1]. Neutral profiles can be calculated with KN1D [2], which uses a collisional-radiative model for the 10 most important electron interactions [3], and elastic collisions are included using a BGK model [4]. The initial objective involved a translation of the original KN1D from IDL to Python, dubbed KN1DPy, adding ease of use and increasing compatibility and coupling with modern use cases. The benchmark of the IDL version is used for KN1DPy with data from C-Mod at MIT. A synthetic database of input/output files from KN1DPy will then be used for machine learning, which will be verified and tweaked with data outside of the training set until an acceptable rate of error is reached. The purpose of ML is to reduce computational stress, which will be particularly valuable for coupling with code that requires a large number of neutral profiles for further analysis and calculations.

[1] S. Mordijck, Nucl. Fusion 60, 082006, 2020

[2] LaBombard B., KN1D: A 1-D Space, 2-D Velocity, Kinetic transport algorithm for atomic and molecular hydrogen in an ionizing plasma, MIT Plasma Science and Fusion Center Report PSFC/RR-01-3; Research Report PSFC/RR-01-3

[3] Sawada, Keiji, and Takashi Fujimoto. "Effective ionization and dissociation rate coefficients of molecular hydrogen in plasma." Journal of applied physics 78.5 (1995): 2913-2924

[4] Bhatnagar, Prabhu Lal, Eugene P. Gross, and Max Krook. "A model for collision processes in gases. I. Small amplitude processes in charged and neutral one-component systems." Physical review 94.3 (1954): 511