skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Fully Convolutional Spatio-Temporal Models for Representation Learning in Plasma Science

Abstract

We have trained a fully convolutional spatio-temporal model for fast and accurate representation learning in the challenging exemplar application area of fusion energy plasma science. The onset of major disruptions is a critically important fusion energy science issue that must be resolved for advanced tokamak plasmas such as the $25B burning plasma international thermonuclear experimental reactor (ITER) experiment. While a variety of statistical methods have been used to address the problem of tokamak disruption prediction and control, recent approaches based on deep learning have proven particularly compelling. In the present paper, we introduce further improvements to the fusion recurrent neural network (FRNN) software suite, which delivered cross-machine disruption predictions with unprecedented accuracy using a large database of experimental signals from two major tokamaks. Up to now, FRNN was based on the long short-term memory (LSTM) variant of recurrent neural networks to leverage the temporal information in the data. Here, we implement and apply the "temporal convolutional neural network (TCN)" architecture to the time-dependent input signals. Furthermore, this allows highly optimized convolution operations to carry the majority of the computational load of training, thus enabling a reduction in training time, and the effective use of high-performance computing resources for hyperparameter tuning.more » At the same time, the TCN-based architecture achieves better predictive performance when compared with the LSTM architecture for various tasks for a representative fusion database.« less

Authors:
 [1];  [2];  [3];  [1];  [1]
  1. Princeton Plasma Physics Lab. (PPPL), Princeton, NJ (United States)
  2. Argonne National Lab. (ANL), Lemont, IL (United States)
  3. Microsoft Corporation, Redmond, WA (United States)
Publication Date:
Research Org.:
Princeton Plasma Physics Lab. (PPPL), Princeton, NJ (United States); General Atomics, San Diego, CA (United States)
Sponsoring Org.:
USDOE Office of Science (SC), Fusion Energy Sciences (FES)
OSTI Identifier:
1814886
Alternate Identifier(s):
OSTI ID: 1830232
Grant/Contract Number:  
AC02-09CH11466; AC05-00OR22725; AC02-05CH11231; AC02-06CH11357; FC02-04ER54698
Resource Type:
Journal Article: Accepted Manuscript
Journal Name:
Journal of Machine Learning for Modeling and Computing
Additional Journal Information:
Journal Volume: 2; Journal Issue: 1; Journal ID: ISSN 2689-3967
Publisher:
Begell House
Country of Publication:
United States
Language:
English
Subject:
97 MATHEMATICS AND COMPUTING; 70 PLASMA PHYSICS AND FUSION TECHNOLOGY; Deep learning; plasma; disruptions; CNN

Citation Formats

Dong, Ge, Felker, Kyle Gerard, Svyatkovskiy, Alexey, Tang, William, and Kates-Harbeck, Julian. Fully Convolutional Spatio-Temporal Models for Representation Learning in Plasma Science. United States: N. p., 2021. Web. doi:10.1615/jmachlearnmodelcomput.2021037052.
Dong, Ge, Felker, Kyle Gerard, Svyatkovskiy, Alexey, Tang, William, & Kates-Harbeck, Julian. Fully Convolutional Spatio-Temporal Models for Representation Learning in Plasma Science. United States. https://doi.org/10.1615/jmachlearnmodelcomput.2021037052
Dong, Ge, Felker, Kyle Gerard, Svyatkovskiy, Alexey, Tang, William, and Kates-Harbeck, Julian. 2021. "Fully Convolutional Spatio-Temporal Models for Representation Learning in Plasma Science". United States. https://doi.org/10.1615/jmachlearnmodelcomput.2021037052.
@article{osti_1814886,
title = {Fully Convolutional Spatio-Temporal Models for Representation Learning in Plasma Science},
author = {Dong, Ge and Felker, Kyle Gerard and Svyatkovskiy, Alexey and Tang, William and Kates-Harbeck, Julian},
abstractNote = {We have trained a fully convolutional spatio-temporal model for fast and accurate representation learning in the challenging exemplar application area of fusion energy plasma science. The onset of major disruptions is a critically important fusion energy science issue that must be resolved for advanced tokamak plasmas such as the $25B burning plasma international thermonuclear experimental reactor (ITER) experiment. While a variety of statistical methods have been used to address the problem of tokamak disruption prediction and control, recent approaches based on deep learning have proven particularly compelling. In the present paper, we introduce further improvements to the fusion recurrent neural network (FRNN) software suite, which delivered cross-machine disruption predictions with unprecedented accuracy using a large database of experimental signals from two major tokamaks. Up to now, FRNN was based on the long short-term memory (LSTM) variant of recurrent neural networks to leverage the temporal information in the data. Here, we implement and apply the "temporal convolutional neural network (TCN)" architecture to the time-dependent input signals. Furthermore, this allows highly optimized convolution operations to carry the majority of the computational load of training, thus enabling a reduction in training time, and the effective use of high-performance computing resources for hyperparameter tuning. At the same time, the TCN-based architecture achieves better predictive performance when compared with the LSTM architecture for various tasks for a representative fusion database.},
doi = {10.1615/jmachlearnmodelcomput.2021037052},
url = {https://www.osti.gov/biblio/1814886}, journal = {Journal of Machine Learning for Modeling and Computing},
issn = {2689-3967},
number = 1,
volume = 2,
place = {United States},
year = {2021},
month = {1}
}

Journal Article:
Free Publicly Available Full Text
This content will become publicly available on January 1, 2022
Publisher's Version of Record

Save / Share: