Prediction of hard x-ray behavior by using the NARX neural network to reduce the destructive effects of runaway electrons in tokamak. (15th November 2021)

• Record Type: Journal Article
• Authors: Alavi, Amir; Saadat, Shervin; Ghanbari, Mohamad Reza; Alavi, Seyed Enayatallah; Kadkhodaie, Ali
• Is Part Of: Physica scripta. Volume 96:Number 12(2021)
• Publication Date: 2021-11-15
• Subjects: methodology -- hard x-ray -- runaway electrons -- NARX network; Physics -- Periodicals; 530.05
• DOI: 10.1088/1402-4896/ac33f7
• Languages: English
• ISSNs: 0031-8949

Abstract

Abstract: The NARX neural network was applied to accurately predict the behavior of Runaway Electrons (REs) in the plasma tokamak. This particular type of artificial neural network was created specifically for time series prediction. The NARX network was built, trained, and tested using inputs from some plasma diagnostic signals (Loop voltage, Hard x-ray, and Plasma current). The network output predicts the time evolution of Hard x-ray (HXR) signals up to 500 μ s, which can be achieved with high accuracy (Mean Absolute Error = 0.003). These results are from experimental data collected during all phases of plasma tokamak discharges. The real-time application of this methodology can pave the way for prompt REs control action. The confinement time increases as the REs decrease, and their destructive effects on the tokamak wall decrease as well. Early prediction of RE behavior is critical in attempting to mitigate their potentially dangerous effects.