Time Series Viewmakers for Robust Disruption Prediction

TL;DR

This study explores the use of a novel time series viewmaker network to generate diverse augmentations or "views" of training data and shows that incorporating views during training improves AUC and F2 scores on DisruptionBench tasks compared to standard or no augmentations.

Abstract

Machine Learning guided data augmentation may support the development of technologies in the physical sciences, such as nuclear fusion tokamaks. Here we endeavor to study the problem of detecting disruptions i.e. plasma instabilities that can cause significant damages, impairing the reliability and efficiency required for their real world viability. Machine learning (ML) prediction models have shown promise in detecting disruptions for specific tokamaks, but they often struggle in generalizing to the diverse characteristics and dynamics of different machines. This limits the effectiveness of ML models across different tokamak designs and operating conditions, which is a critical barrier to scaling fusion technology. Given the success of data augmentation in improving model robustness and generalizability in other fields, this study explores the use of a novel time series viewmaker network to generate diverse augmentations or "views" of training data. Our results show that incorporating views during training improves AUC and F2 scores on DisruptionBench tasks compared to standard or no augmentations. This approach represents a promising step towards developing more broadly applicable ML models for disruption avoidance, which is essential for advancing fusion technology and, ultimately, addressing climate change through reliable and sustainable energy production.

Figures (6)

• Figure 1: Overview of the time series viewmaker. $V_t$ and $V_s$ are generator networks which generate perturbed time series from $X_t$ and $X_s$ respectively.
• Figure 2: Illustration of the categorization of true and false positives, and true and false negatives. Figure adapted from ref. keith2024risk.
• Figure 3: A UMAP clustering of disruptive discharges from each machine. Many discharges from C-MOD exhibit distinct behavior from those from DIII-D and EAST. Thus, to challenge models in their ability to learn general disruptive features, we designate C-MOD as the "new machine" parameter in our experiments.
• Figure 4: Comparison of dynamic time warping (DTW) similarities between disruptive discharges and their views (in cyan) and their tsaugs (in orange). Discharges were sampled evenly at random across each machine.
• Figure 5: A comparison of four discharges from Alcator C-Mod for unrolled disruptivity on the test set, chosen based on outcomes of an LSTMFormer trained of views. We annotate the chart with "X"'s to show the points at which the disruption models would have triggered the DMS. The disruptivity metric, on the y-axis, is unitless. Finally, note that the timebases are not true, but relative to the start of the recorded data.