A Python Version of KN1D for the Training of a Neutral Density Predicting Machine Learning Algorithm

In fusion research, predicting neutral profiles is critical to understanding the fueling requirements for a fusion reactor [1]. KN1D [2] uses a collisional-radiative model for the 10 most important electron interactions [3], and elastic collisions are included using a BGK model [4]. We will first convert KN1D into Python to simplify its use for the wider community, improve compatibility with experimental data and predictive modeling, and allow easier coupling to ML and AI algorithms. The Python version is verified by comparing it to the previous IDL version using data from the C-Mod tokamak at MIT. A ML algorithm will then be trained with a synthetic database of input/output files from KN1D. Inputs outside of the training data are then put through both the algorithm and KN1D, with any deviations being amended after comparing outputs. The main advantage of a well-trained machine learning algorithm is that it can significantly reduce the computational requirements of calculating neutral densities compared to KN1D.

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

[2] B. LaBombard, KN1D, MIT PSFC/RR-01-3

[3] K. Sawada and T. Fujimoto. Journal of applied physics 78.5 (1995): 2913-2924

[4] P. L. Bhatnagar, E. P. Gross, and M. Krook. Physical review 94.3 (1954): 511