Physics-augmented surrogate model for linear instability analysis of high- n toroidal Alfvén eigenmodes in tokamaks

Abstract

We propose the physics-augmented surrogate model for linear instability analysis of high- n toroidal Alfvén eigenmodes (TAEs) in tokamaks. The database for high- n TAE linear instabilities is generated by the global initial-value simulations by Gyrokinetic magnetohydrodynamic Energetic-particle Code (GMEC), and also the local eigenvalue code Energetic-particle TAE Eigenvalue Code (ETEC). The equations solved by ETEC are much simplified from the first-principle equations solved by GMEC, with high- n ballooning representation. The local solutions from ETEC exhibit a strong correlation with the global solutions from GMEC. The multi-layer perceptron based surrogate models are trained, taking local normalized equilibrium parameters as inputs, and the real frequency and growth rate of TAEs obtained from GMEC as outputs. Our results indicate that incorporating ETEC results as prior physical information into the inputs leads to a substantial improvement in the generalization performance of the surrogate model.