Towards Transparent and Accurate Plasma State Monitoring at JET

TL;DR

This work addresses the challenge of transparent and accurate plasma state monitoring in tokamaks by integrating supervised disruption prediction with unsupervised latent-dynamics learning in a multitask transformer framework. The model maps high-dimensional plasma measurements to a compact 2D latent space (L=2), trained with four tasks and a latent-prior regularizer to promote smooth dynamics and interpretability. Disruption-prediction results are competitive, with zero missed alarms and high success rates, and the latent space reveals physically meaningful structures corresponding to disruption mechanisms and precursors. The approach provides actionable triggers for control mode switching and enables qualitative and quantitative exploration of plasma-state evolution, with potential extensions in self-supervised pretraining and higher-frequency data for improved scenario optimization.

Abstract

Controlling and monitoring plasma within a tokamak device is complex and challenging. Plasma off-normal events, such as disruptions, are hindering steady-state operation. For large devices, they can even endanger the machine's integrity and it represents in general one of the most serious concerns for the exploitation of the tokamak concept for future power plants. Effective plasma state monitoring carries the potential to enable an understanding of such phenomena and their evolution which is crucial for the successful operation of tokamaks. This paper presents the application of a transparent and data-driven methodology to monitor the plasma state in a tokamak. Compared to previous studies in the field, supervised and unsupervised learning techniques are combined. The dataset consisted of 520 expert-validated discharges from JET. The goal was to provide an interpretable plasma state representation for the JET operational space by leveraging multi-task learning for the first time in the context of plasma state monitoring. When evaluated as disruption predictors, a sequence-based approach showed significant improvements compared to the state-based models. The best resulting network achieved a promising cross-validated success rate when combined with a physical indicator and accounting for nearby instabilities. Qualitative evaluations of the learned latent space uncovered operational and disruptive regions as well as patterns related to learned dynamics and global feature importance. The applied methodology provides novel possibilities for the definition of triggers to switch between different control scenarios, data analysis, and learning as well as exploring latent dynamics for plasma state monitoring. It also showed promising quantitative and qualitative results with warning times suitable for avoidance purposes and distributions that are consistent with known physical mechanisms.

Figures (7)

• Figure 1: Sequence-based architecture of the employed model leveraging multi-task learning. By using auto-regressive attention masks the transformer encoder ensures representation and prediction causality.
• Figure 2: Instabilities found in a fixed window of $\pm200$ms around false alarms of sequence-based models.
• Figure 3: Warning times for different detection methods using sequence-based models (a) on training discharges and (b) on unseen test discharges.
• Figure 4: Visualization of the low dimensional projection of the training data. In (a) the color scale is log scaled and indicates the time to boundary quantity. For (b) the color scales indicate the predicted state disruptivity and the time variable of the shown discharge.
• Figure 5: (a) Multiple unseen discharges projected on the learned latent space where their start is called head and their end tail. (b) Visualization of all heads and tails of the training data.