Disruption Prediction on JET during the ILW Experimental Campaigns

Investigation of Machine Learning Techniques for Disruption Prediction Using JET Data

Disruption prediction and mitigation is of key importance in the development of sustainable tokamak reactors. Machine learning has become a key tool in this endeavour. In this paper, multiple machine learning models are tested and compared. A focus has been placed on the analysis of a transition to dimensionless input quantities. The methods used in this paper are the support vector machine, two-tiered support vector machine, random forest, gradient-boosted trees and long-short term memory. The performance between different models is remarkably similar, with the support vector machine attaining a slightly better accuracy score. The similarity could indicate issues with the dataset, but further study is required to confirm this. Both the two-tiered model and long-short term memory performed below expectations. The former could be attributed to an implementation which did not allow error propagation between tiers. The latter could be attributed to high noise and low frequency of the input signals. Dimensionless models experienced an expected decrease in performance, caused by a loss of information in the conversion. However, random forest and gradient boosted trees experienced a significantly lower decrease, making them more suitable for dimensionless predictors. From the disruption detection times, it was concluded that several disruptions could be predicted at more than 600ms in advance. A feature importance study using the random forest indicated the negative impact of high noise and missing data in the database, suggesting improvements in data preparation for future work and the potential reevaluation of some of the selected portable features due to poor performance.

Explainable deep learning for the analysis of MHD spectrograms in nuclear fusion

In the nuclear fusion community, there are many specialized techniques to analyze the data coming from a variety of diagnostics. One of such techniques is the use of spectrograms to analyze the magnetohydrodynamic (MHD) behavior of fusion plasmas. Physicists look at the spectrogram to identify the oscillation modes of the plasma, and to study instabilities that may lead to plasma disruptions. One of the major causes of disruptions occurs when an oscillation mode interacts with the wall, stops rotating, and becomes a locked mode. In this work, we use deep learning to predict the occurrence of locked modes from MHD spectrograms. In particular, we use a convolutional neural network with class activation mapping to pinpoint the exact behavior that the model thinks is responsible for the locked mode. Surprisingly, we find that, in general, the model explanation agrees quite well with the physical interpretation of the behavior observed in the spectrogram.

Predicting disruptive instabilities in controlled fusion plasmas through deep learning

Nuclear fusion power delivered by magnetic-confinement tokamak reactors holds the promise of sustainable and clean energy¹. The avoidance of large-scale plasma instabilities called disruptions within these reactors^2,3 is one of the most pressing challenges^4,5, because disruptions can halt power production and damage key components. Disruptions are particularly harmful for large burning-plasma systems such as the multibillion-dollar International Thermonuclear Experimental Reactor (ITER) project⁶ currently under construction, which aims to be the first reactor that produces more power from fusion than is injected to heat the plasma. Here we present a method based on deep learning for forecasting disruptions. Our method extends considerably the capabilities of previous strategies such as first-principles-based⁵ and classical machine-learning^7-11 approaches. In particular, it delivers reliable predictions for machines other than the one on which it was trained-a crucial requirement for future large reactors that cannot afford training disruptions. Our approach takes advantage of high-dimensional training data to boost predictive performance while also engaging supercomputing resources at the largest scale to improve accuracy and speed. Trained on experimental data from the largest tokamaks in the United States (DIII-D¹²) and the world (Joint European Torus, JET¹³), our method can also be applied to specific tasks such as prediction with long warning times: this opens up the possibility of moving from passive disruption prediction to active reactor control and optimization. These initial results illustrate the potential for deep learning to accelerate progress in fusion-energy science and, more generally, in the understanding and prediction of complex physical systems.