An advanced plasma current tomography method based on Bayesian inference and neural networks for real-time application

Abstract

An advanced plasma current tomography method is established for the Experimental Advanced Superconducting Tokamak (EAST), which combines Bayesian probability theory and neural networks. It is different from the existing current tomography method based on a conditional autoregressive (CAR) prior. Specifically, the CAR prior is replaced with an advanced squared exponential (ASE) kernel function prior. Therefore, the proposed method can overcome the deficiencies of the CAR prior, where the calculated core current is lower than the reference current and the uncertainty becomes severe after introducing noise in the diagnostics. The ASE kernel prior is developed from the squared exponential kernel function by integrating the useful information from the reference discharge. The ASE kernel prior adopts nonstationary hyperparameters and introduces the current profile into the hyperparameters, which can make the shape of the current profile more flexible in space. To provide a suitable reference discharge, a neural network model is also trained. The execution time is less than 1 ms for each time slice, which indicates its potential for application in future real-time plasma feedback control.