Hybrid deep-learning architecture for general disruption prediction across multiple tokamaks

Abstract

In this paper, we present a new deep-learning disruption-prediction algorithm based on important findings from explorative data analysis which effectively allows knowledge transfer from existing devices to new ones, thereby predicting disruptions using very limited disruption data from the new devices. The explorative data analysis, conducted via unsupervised clustering techniques confirms that time-sequence data are much better separators of disruptive and non-disruptive behavior than the instantaneous plasma-state data, with further advantageous implications for a sequence-based predictor. Based on such important findings, we have designed a new algorithm for multi-machine disruption prediction that achieves high predictive accuracy for the C-Mod (AUC = 0.801), DIII-D (AUC = 0.947) and EAST (AUC = 0.973) tokamaks with limited hyperparameter tuning. Through numerical experiments, we show that a boosted accuracy (AUC = 0.959) is achieved for the EAST predictions by including only 20 disruptive discharges with thousands of non-disruptive discharges from EAST in the training, combined with more than a thousand discharges from DIII-D and C-Mod. The improvement in the predictive ability obtained by combining disruption data from other devices is found to be true for all permutations of the three devices. Furthermore, by comparing the predictive performance of each individual numerical experiment, we find that non-disruption data are machine-specific, while disruption data from multiple devices contain device-independent knowledge that can be used to inform predictions for disruptions occurring in a new device.

Disruption prediction with artificial intelligence techniques in tokamak plasmas

“…First, they are not derived from first principles but are empirical. This means that their results are difficult to interpret in terms of plasma dynamics, and whether they can be extrapolated to future, larger devices 38,39 remains unclear. Second, traditional machine-learning predictors require very large amounts of data for the training.…”

2022 Review of Data-Driven Plasma Science

“…Deep-learning algorithm for multi-machine disruption prediction has been proposed and achieved high predicting accuracy across multiple tokamaks [421]. This means device-independent representations of disruptive characteristics have been identified.…”

“…The approaches of interpretable machine learning models, which are contrast methodology of deep learning [421], neural network [420] and generative topographic mapping [422], are attracting interests because not only they improve prediction capability but also their resultant expression enables exploration of underlying disruption physics. Physics validation of the model/hypothesis would secure limitation of generalization performance.…”

Disruption prediction based on fusion feature extractor on J-TEXT

“…[15] The hybrid deep-learning (HDL) disruption-prediction framework is inspired by the work in machine translation. [18] Though performance of the model could be improved, models it referred to are designed for the specific tasks in other fields, which are able to better extract features in their own fields rather than disruption prediction. Few models have been designed specifically for disruption prediction or for tokamak diagnostics feature extracting.…”