Inversion of the fast ion 1D velocity distribution from collective Thomson scattering spectra with machine learning methods

Fusion energy holds great potential as a sustainable and environmentally friendly energy source. To characterize the fast ion distribution in future fusion devices, collective Thomson scattering (CTS) diagnostic has emerged as crucial tools. In this study, an inference approach based on machine learning method has been developed to inverse the fast-ion 1D velocity distribution from CTS spectra. Specifically, we construct a Support Vector Regression (SVR) model with a Radial Basis Function kernel. Through proper implementation, the SVR inversion model achieves an impressive accuracy of 1.81% uniform MAPE at the relative noise scale of 0.3. Afterwards, the conventional inversion method employed by Least Squares with Nuisance Parameter (LSN) was compared with our inversion method, and superior performance was demonstrated by our approach. Overall, CTS inversion method uses SVR demonstrate favorable outcomes and noise robustness.