Deep convolutional neural networks for multi-scale time-series classification and application to disruption prediction in fusion devices

Churchill, R. M. ; Tobias, B.; Zhu, Y.
Issue date: 2019
Rights:
Creative Commons Attribution 4.0 International (CC BY)
Cite as:
Churchill, R. M., Tobias, B., & Zhu, Y. Deep convolutional neural networks for multi-scale time-series classification and application to disruption prediction in fusion devices [Data set]. Princeton Plasma Physics Laboratory, Princeton University. https://doi.org/10.11578/1661171
@electronic{churchill_r_m_unknown,
  author      = {Churchill, R. M. and
                Tobias, B. and
                Zhu, Y.},
  title       = {{Deep convolutional neural networks for m
                ulti-scale time-series classification an
                d application to disruption prediction i
                n fusion devices}},
  publisher   = {{Princeton Plasma Physics Laboratory, Pri
                nceton University}},
  url         = {https://doi.org/10.11578/1661171}
}
Description:

The multi-scale, mutli-physics nature of fusion plasmas makes predicting plasma events challenging. Recent advances in deep convolutional neural network architectures (CNN) utilizing dilated convolutions enable accurate predictions on sequences which have long-range, multi-scale characteristics, such as the time-series generated by diagnostic instruments observing fusion plasmas. Here we apply this neural network architecture to the popular problem of disruption prediction in fusion tokamaks, utilizing raw data from a single diagnostic, the Electron Cyclotron Emission imaging (ECEi) diagnostic from the DIII-D tokamak. ECEi measures a fundamental plasma quantity (electron temperature) with high temporal resolution over the entire plasma discharge, making it sensitive to a number of potential pre-disruptions markers with different temporal and spatial scales. Promising, initial disruption prediction results are obtained training a deep CNN with large receptive field ({$\sim$}30k), achieving an $F_1$-score of {$\sim$}91\% on individual time-slices using only the ECEi data.

Show More
# Filename Filesize
1 README.txt 757 Bytes
2 ARK_DATA.zip 4.38 MB