Approximation of Daily AMS‐02 Spectra With Machine Learning Methods

Abstract Almost three thousand daily AMS‐02 proton spectra from 2011 to 2019 offer the most precise and extensive data set of cosmic ray spectra covering a wide energy range. As such, they offer a unique opportunity to test machine learning algorithms for approximating cosmic ray proton spectra based on the inputs usually available to solar modulation models. We evaluated how various machine learning techniques approximate the temporal evolution of the AMS‐02 flux for a wide range of published rigidity bins from 2011 to 2019, primarily focusing on the feasibility and effectiveness of machine learning approaches compared to the traditional force field model in approximating cosmic ray proton spectra. The machine learning methods are very accurate, particularly in comparison to the established force field approach, and significantly improve the approximation of the behavior of cosmic rays.