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

Title: Predicting disruptive instabilities in controlled fusion plasmas through deep learning

Journal Article · · Nature (London)
 [1];  [2];  [3]
  1. Harvard Univ., Cambridge, MA (United States); Princeton Plasma Physics Lab. (PPPL), Princeton, NJ (United States)
  2. Princeton Univ., NJ (United States)
  3. Princeton Plasma Physics Lab. (PPPL), Princeton, NJ (United States); Princeton Univ., NJ (United States)

Nuclear fusion power via magnetic confinement tokamak reactors carries the promise of sustainable and clean energy production for the future [1]. The avoidance of large-scale plasma instabilities called disruptions [2, 3] is one of the most pressing challenges [4, 5] towards this goal. Disruptions are particularly deleterious for large reactors such as the multi-billion dollar international ITER project [6] currently under construction, whose reaction for the first time aims to produce more power from fusion than is injected to heat the plasma. Here, we present a new approach to forecast disruptions based on deep learning that for the first time overcomes the crucial limitations of past work, such as first-principles based [5] and classical machine learning approaches [7, 8, 9, 10, 11]. In particular, our method for the first time (i) delivers reliable predictions on machines other than the one it was trained on — a crucial requirement for large future reactors that cannot afford “training” disruptions; (ii) utilizes high-dimensional training data such as profiles to add new physics information and boost predictive performance; and (iii) engages supercomputing at the largest scale to deliver solutions with improved accuracy and speed. Trained on experimental data from the largest tokamaks in the US (DIII-D [12]) and the world (JET [13]), our method moreover can be tuned for physics-specific tasks such as prediction with long warning times, and opens up possible avenues for moving forward the goal from passive disruption prediction to active reactor control and optimization. These results illustrate the potential for deep learning to accelerate progress in fusion energy science and in the understanding and prediction of complex systems in the physical sciences in general.

Research Organization:
Princeton Plasma Physics Laboratory (PPPL), Princeton, NJ (United States); General Atomics, San Diego, CA (United States)
Sponsoring Organization:
USDOE Office of Science (SC), Fusion Energy Sciences (FES)
Grant/Contract Number:
FG02-97ER25308; FC02-04ER54698
Alternate ID(s):
OSTI ID: 1566845
Journal Information:
Nature (London), Vol. 568, Issue 7753; ISSN 0028-0836
Nature Publishing GroupCopyright Statement
Country of Publication:
United States
Citation Metrics:
Cited by: 133 works
Citation information provided by
Web of Science

References (42)

Commissioning of electron cyclotron emission imaging instrument on the DIII-D tokamak and first data journal October 2010
Statistical analysis of disruptions in JET journal April 2009
Deep learning journal May 2015
Fast visible camera installation and operation in JET conference January 2008
Advances and challenges in computational plasma science journal January 2005
Human-level control through deep reinforcement learning journal February 2015
Accelerating recurrent neural network training using sequence bucketing and multi-GPU data parallelization conference August 2016
Survey of disruption causes at JET journal April 2011
Cost-sensitive learning by cost-proportionate example weighting conference January 2003
Simultaneous closed-loop control of the current profile and the electron temperature profile in the TCV tokamak conference July 2015
Disruption Prediction on JET during the ILW Experimental Campaigns journal April 2016
Disruption mitigation studies in DIII-D journal May 1999
Real time equilibrium reconstruction for tokamak discharge control journal July 1998
JET ITER-like wall—overview and experimental programme journal December 2011
Overview of the JET results in support to ITER journal June 2017
Prototype of an adaptive disruption predictor for JET based on fuzzy logic and regression trees journal February 2008
Training distributed deep recurrent neural networks with mixed precision on GPU clusters conference January 2017
Disruption forecasting at JET using neural networks journal December 2003
Improved feature selection based on genetic algorithms for real time disruption prediction on JET journal September 2012
A design retrospective of the DIII-D tokamak journal May 2002
ImageNet: A large-scale hierarchical image database
  • Deng, Jia; Dong, Wei; Socher, Richard
  • 2009 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPR Workshops), 2009 IEEE Conference on Computer Vision and Pattern Recognition
conference June 2009
The use of the area under the ROC curve in the evaluation of machine learning algorithms journal July 1997
Disruptions in tokamaks journal November 1995
The ITER design journal April 2002
XGBoost: A Scalable Tree Boosting System conference January 2016
Cyclical Learning Rates for Training Neural Networks conference March 2017
Results of the JET real-time disruption predictor in the ITER-like wall campaigns journal October 2013
Disruptions in ITER and strategies for their control and mitigation journal August 2015
Requirements for Triggering the ITER Disruption Mitigation System journal April 2016
Tokamak disruption alarm based on a neural network model of the high- beta limit journal June 1997
Feature extraction for improved disruption prediction analysis at JET journal October 2008
The Power of an Idea: The International Impacts of the Grand Challenges for Engineering journal March 2016
Solving the quantum many-body problem with artificial neural networks journal February 2017
Machine learning phases of matter journal February 2017
A cross-tokamak neural network disruption predictor for the JET and ASDEX Upgrade tokamaks journal April 2005
Random forest-based prediction of stroke outcome journal May 2021
Plasma Physics and Fusion Energy journal November 2008
Plasma Physics and Fusion Energy journal November 2008
Deep Learning text January 2018
Cyclical Learning Rates for Training Neural Networks preprint January 2015
Machine learning phases of matter text January 2016
Accelerating recurrent neural network training using sequence bucketing and multi-GPU data parallelization text January 2017

Cited By (13)

Machine learning control for disruption and tearing mode avoidance journal February 2020
Progress in disruption prevention for ITER journal June 2019
A machine learning approach based on generative topographic mapping for disruption prevention and avoidance at JET journal August 2019
Disruptive plasma simulations in EAST including 3D effects journal September 2019
Deep neural network Grad–Shafranov solver constrained with measured magnetic signals journal December 2019
A linear equation based on signal increments to predict disruptive behaviours and the time to disruption on JET journal December 2019
Machine learning and big scientific data
  • Hey, Tony; Butler, Keith; Jackson, Sam
  • Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, Vol. 378, Issue 2166
journal January 2020
“Zhores” — Petaflops supercomputer for data-driven modeling, machine learning and artificial intelligence installed in Skolkovo Institute of Science and Technology journal October 2019
Deep learning based surrogate models for first-principles global simulations of fusion plasmas journal November 2021
Deep convolutional neural networks for multi-scale time-series classification and application to disruption prediction in fusion devices text January 2019
Offline Contextual Bayesian Optimization for Nuclear Fusion preprint January 2020
Using LSTM for the Prediction of Disruption in ADITYA Tokamak text January 2020
On the Universal Transformation of Data-Driven Models to Control Systems text January 2021