Supervised learning approaches to modeling pedestal density

Pedestal is the key to conventional high performance plasma scenarios in tokamaks. However, fidelity simulations of pedestal plasmas are extremely challenging due multiple physical processes and scales that encompassed by tokamak pedestals. The leading paradigm for predicting top pressure EPED-like models. EPED does not predict density, $n_\text{e,ped}$, but requires it as an input. EUROPED employs simplified models, such log-linear regression, constrain $n_\text{e,ped}$ with machine control parameters model. these models often show disagreements experimental observations do use all available numerical categorical information. In this work observed using same input parameters, decision tree ensembles deep learning improve predictive quality about 23% relative obtained scaling laws, measured root mean square error. Including both categorical, leads further improvement 13%. Finally, was tested when including global normalized effective charge state inputs, known impact Surprisingly, lead only a few percent quality.