Neural-Network Model for Linear MHD Stability Analysis of Tokamak Edge Pedestals

The new neural-network model for the pedestal linear MHD (magnetohydrodynamic) stability analysis is developed, to accelerate analysis speed and to reduce numerical burden. This model predicts linear growth rates of edge-localized MHD instabilities for KSTAR-like (Korea Superconducting Tokamak Advanced Research) plasma with a pedestal structure at the edge. The model is trained by a data set consisting of numerical parametric plasma equilibria and stability calculation results. It has successfully predicted the growth rates and the most unstable toroidal mode number for benchmark cases within reasonable errors. Required time for conducting peeling-ballooning stability theory-based pedestal stability analysis has been reduced from few hours to seconds by the model. A few ideas are also suggested for the upgrade of the model.