A new vertical instability predictor via precursor oscillation detection with performance monitoring of equilibrium controller

Abstract

Abstract We first propose an accurate and robust vertical instability predictor by using a support vector machine (SVM), one of the machine learning methods. The predictor is trained to detect precursor oscillation by using newly introduced classification parameters to measure the equilibrium controller performance, which is obtained by the adaptive voltage allocation scheme (Inoue et al 2021 Nuclear Fusion 61 096009). Furthermore, multi-layered preprocessing filters are newly introduced for the SVM training/prediction, which improves the prediction accuracy under highly imbalanced conditions, where ∼ 500 disruptive data while ∼ 3 × 1 0 6 non-disruptive data. The classification parameters can be calculated only by the current centroid, which suggests that the proposed predictor is robust against the extrapolation for the experiment and will be validated in JT-60SA experiments.