# Investigation of Machine Learning Techniques for Disruption Prediction Using JET Data

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Data

## 3. Methods

#### 3.1. Support Vector Machines

#### 3.2. Random Forest

#### 3.3. Gradient-Boosted Trees

#### 3.4. Long Short-Term Memory

## 4. Results

#### 4.1. Predictor Performance

#### 4.2. Detection Time and Cumulative Detection Rate

#### 4.3. Relative Feature Importance

#### 4.4. Computational Cost

## 5. Conclusions

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## References

- Boozer, A.H. Theory of tokamak disruptions. Phys. Plasmas
**2012**, 19, 058101. [Google Scholar] [CrossRef] - Cardella, A.; Gorenflo, H.; Lodato, A.; Ioki, K.; Raffray, R. Effects of plasma disruption events on iter first wall materials. J. Nucl. Mater.
**2000**, 283–287, 1105–1110. [Google Scholar] [CrossRef] - Vega, J.; Citrin, J.; Ho, A.; Huijsmans, G.T.; Marin, M.; van de Plassche, K.L. Disruption prediction with artificial intelligence techniques in tokamak plasmas. Nat. Phys.
**2022**, 18, 741–750. [Google Scholar] [CrossRef] - Lehnen, M.; Maruyama, S. Iter Disruption Mitigation Workshop; Technical Report, ITR-18-002; ITER Organisation: Saint-Paul-lez-Durance, France, 2017. [Google Scholar]
- De Vries, P.; Johnson, M.; Alper, B.; Buratti, P.; Hender, T.; Koslowski, H.; Riccardo, V. Survey of disruption causes at jet. Nucl. Fusion
**2011**, 5, 053018. [Google Scholar] [CrossRef] - Baylor, L.; Combs, S.; Duckworth, R.; Lyttle, M.; Meitner, S.; Rasmussen, D.A.; Maruyama, S. Pellet injection technology and its application on iter. IEEE Trans. Plasma Sci.
**2016**, 44, 1489–1495. [Google Scholar] [CrossRef] - Hollmann, E.M.; Aleynikov, P.B.; Fulop, T.; Humphreys, D.A.; Izzo, V.A.; Lehnen, M.; Lukash, V.E.; Papp, G.; Pautasso, G.; Saint-Laurent, F.; et al. Status of research toward the iter disruption mitigation system. Phys. Plasmas
**2015**, 22, 21802. [Google Scholar] [CrossRef] - Jachmich, S.; Kruezi, U.; Card, P.; Deakin, K.; Kinna, D.; Koslowski, H.; Lambertz, H.; Lehnen, M. Implementation of a new disruption mitigation system into the control system of jet. Fusion Eng. Des.
**2015**, 96–97, 633–636. [Google Scholar] [CrossRef] - Moreno, R.; Vega, J.; Dormido-Canto, S.; Pereira, A.; Murari, A.; Contributors JET. Disruption prediction on jet during the ilw experimental campaigns. Fusion Sci. Technol.
**2016**, 69, 485–494. [Google Scholar] [CrossRef] - Rattá, G.A.; Vega, J.; Murari, A.; Vagliasindi, G.; Johnson, M.F.; de Vries, P.C. An advanced disruption predictor for jet tested in a simulated real-time environment. Nucl. Fusion
**2010**, 50, 025005. [Google Scholar] [CrossRef] - Cannas, B.; Fanni, A.; Pautasso, G.; Sias, G.; Sonato, P. An adaptive real-time disruption predictor for ASDEX upgrade. Nucl. Fusion
**2010**, 50, 075004. [Google Scholar] [CrossRef] - Zheng, W.; Hu, F.; Zhang, M.; Chen, Z.; Zhao, X.; Wang, X.; Shi, P.; Zhang, X.; Zhang, X.; Zhou, Y.; et al. Hybrid neural network for density limit disruption prediction and avoidance on j-text tokamak. Nucl. Fusion
**2018**, 58, 056016. [Google Scholar] [CrossRef] - Rea, C.; Granetz, R.; Montes, K.; Tinguely, R.; Eidietis, N.; Hanson, J.; Sammuli, B. Disruption prediction investigations using machine learning tools on diii-d and alcator c-mod. Plasma Phys. Control. Fusion
**2018**, 60, 084004. [Google Scholar] [CrossRef] - Guo, B.H.; Chen, D.L.; Shen, B.; Rea, C.; Granetz, R.S.; Zeng, L.; Hu, W.H.; Qian, J.P.; Sun, Y.W.; Xiao, B.J. Disruption prediction on east tokamak using a deep learning algorithm. Plasma Phys. Control. Fusion.
**2021**, 63, 115007. [Google Scholar] [CrossRef] - Kates-Harbeck, J.; Svyatkovskiy, A.; Tang, W. Predicting disruptive instabilities in controlled fusion plasmas through deep learning. Nature
**2019**, 568, 526–531. [Google Scholar] [CrossRef] - Pau, A.; Fanni, A.; Carcangiu, S.; Cannas, B.; Sias, G.; Murari, A.; Rimini, F. A machine learning approach based on generative topographic mapping for disruption prevention and avoidance at JET. Nucl. Fusion
**2019**, 59, 106017. [Google Scholar] [CrossRef] - Zhu, J.X.; Rea, C.; Montes, K.; Granetz, R.S.; Sweeney, R.; Tinguely, R.A. Hybrid deep learning architecture for general disruption prediction across tokamaks. Nucl. Fusion
**2021**, 61, 026007. [Google Scholar] [CrossRef] - Croonen, J. Predicting Plasma Disruptions in Tokamak Fusion Reactors Using Machine Learning. Master’s Thesis, KU Leuven, Leuven, Belgium, 2019. [Google Scholar]
- Cortes, C.; Vapnik, V. Support-vector networks. Mach. Learn.
**1995**, 20, 273–297. [Google Scholar] [CrossRef] - Breiman, L. Random forests. Mach. Learn.
**2001**, 45, 5–32. [Google Scholar] [CrossRef] - Friedman, J. Greedy function approximation: A gradient boosting machine. Ann. Stat.
**2001**, 29, 1189–1232. [Google Scholar] [CrossRef] - Hochreiter, S.; Schmidhuber, J. Long short-term memory. Neural Comput.
**1997**, 9, 1735–1780. [Google Scholar] [CrossRef] - Eidietis, N.; Gerhardt, S.; Granetz, R.; Kawano, Y.; Lehnen, M.; Lister, J.; Pautasso, G.; Riccardo, V.; Tanna, R.; Thornton, A. The ITPA disruption database. Nucl. Fusion
**2015**, 55, 063030. [Google Scholar] [CrossRef] - Rattá, G.; Vega, J.; Murari, A. Improved feature selection based on genetic algorithms for real time disruption prediction on jet. Fusion Eng. Des.
**2012**, 87, 1670–1678. [Google Scholar] [CrossRef] - Pustovitov, V.D.; Ryabushev, E.A. Effect of pressure anisotropy on diamagnetic signal in a tokamak with noncircular plasma cross section. Plasma Phys. Rep.
**2021**, 47, 947–955. [Google Scholar] [CrossRef] - Schlisio, G. Analysis of the Gas Balance for Wendelstein 7-x; Technical Report; University of Greifswald: Greifswald, Germany, 2020; p. 9. [Google Scholar]
- Aledda, R.; Cannas, B.; Fanni, A.; Pau, A.; Sias, G. Improvements in disruption prediction at asdex upgrade. Fusion Eng. Des.
**2015**, 96–97, 698–702. [Google Scholar] [CrossRef] - Tang, W.; Parsons, M.; Feigbush, E.; Choi, J.; Kurc, T. Big data machine learning for disruption prediction. In Proceedings of the 26th IAEA Fusion Energy Conference, Kyoto, Japan, 17–22 October 2016. [Google Scholar]
- Windsor, C.; Pautasso, G.; Tichmann, C.; Buttery, R.; Hende, T.; JET EFDA Contributors the ASDEX Upgrade Team. A cross-tokamak neural network disruption predictor for the JET and ASDEX upgrade tokamaks. Nucl. Fusion
**2005**, 45, 337–350. [Google Scholar] [CrossRef] - Hastie, T.; Tibshirani, R.; Friedman, J. The Elements of Statistical Learning, 2nd ed.; Springer: Berlin/Heidelberg, Germany, 2009. [Google Scholar]
- Pedregosa, F. Scikit-learn: Machine learning in Python. J. Mach. Learn. Res.
**2011**, 12, 2825–2830. [Google Scholar] - Goodfellow, I.; Bengio, Y.; Courville, A. Deep Learning; The MIT Press: Cambridge, MA, USA, 2016. [Google Scholar]
- Paszke, A. Pytorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems 32; Curran Associates, Inc.: Sydney, NSW, Australia, 2019; pp. 8024–8035. Available online: http://papers.neurips.cc/paper/9015-pytorch-an-imperative-style-high-performance-deep-learning-library.pdf (accessed on 4 December 2022).
- Kingma, D.; Ba, J. Adam: A method for stochastic optimization. In Proceedings of the International Conference on Learning Representations, Banff, AB, Canada, 14–16 April 2014; p. 12. [Google Scholar]
- Moreno, R.; Vega, J.; Murari, A.; Dormido-Canto, S.; López, J.M.; Ramírez, J.M. Robustness and increased time resolution of jet advanced predictor of disruptions. Plasma Phys. Control. Fusion
**2014**, 56, 114003. [Google Scholar] [CrossRef] - Rattá, G.A.; Vega, J.; Murari, A. Simulation and real-time replacement of missing plasma signals for disruption prediction: An implementation with apodis. Plasma Phys. Control. Fusion
**2014**, 56, 114004. [Google Scholar] [CrossRef] - Lin, W.C.; Tsai, C.F. Missing value imputation: A review and analysis of the literature (2006–2017). Artif. Intell. Rev.
**2020**, 53, 1487–1509. [Google Scholar] [CrossRef] - López, J.M.; Vega, J.; Alves, D.; Dormido-Canto, S.; Murari, A.; Ramírez, J.M.; Felton, R.; Ruiz, M.; de Arcas, G. Implementation of the disruption predictor apodis in jet’s real-time network using the marte framework. IEEE Trans. Nucl. Sci.
**2014**, 61, 741–744. [Google Scholar] [CrossRef] - Louppe, G.; Wehenkel, L.; Sutera, A.; Geurts, P. Understanding variable importances in forests of randomized trees. In Proceedings of the Advances in Neural Information Processing Systems, Lake Tahoe, NV, USA, 5–10 December 2013; Volume 26. Available online: https://proceedings.neurips.cc (accessed on 4 December 2022).

**Figure 1.**Relative importance of the different standard input features, as obtained from the random forest model.

**Figure 2.**Schematic of the sliding window scheme used by the 2-tiered SVM. (Source: Croonen [18]).

**Figure 3.**Schematic of the sliding many-to-one implementation of LSTM. (Source: Croonen [18]).

**Figure 5.**Example of diamagnetic energy and its time derivative, showing the noise on the signals. The vertical line is the moment of disruption.

**Figure 9.**Relative importance of the different dimensionless input features, as obtained from the random forest model.

Name | Description | Data Availability | $\frac{\mathit{\sigma}}{\mathit{\mu}}$ |
---|---|---|---|

${I}_{pla}$ | Plasma current $\left[A\right]$ | 100% | 0.83 |

$MLA$ | Mode lock amplitude $\left[T\right]$ | 100% | 2.19 |

${l}_{i}$ | Plasma internal inductance | 81.3% | 0.90 |

${W}_{dia}$ | Diamagnetic energy $\left[J\right]$ | 99.0% | 1.35 |

$\dot{{W}_{dia}}$ | Time derivative of the diamagnetic energy $\left[W\right]$ | 99.0% | 64.37 |

${n}_{e}$ | Electron density $\left[{m}^{-3}\right]$ | 99.7% | 1.17 |

${P}_{out}$ | Radiated output power $\left[W\right]$ | 99.7% | 1.80 |

${P}_{NBI}$ | Neutral beam injection input power $\left[W\right]$ | 70.0% | 1.75 |

${P}_{ICRH}$ | Ion cyclotron radio heating input power $\left[W\right]$ | 42.5% | 3.05 |

${q}_{95}$ | Edge safety factor | 100% | 1.08 |

${B}_{\varphi}$ | Toroidal magnetic field strength $\left[T\right]$ | 100% | 0.81 |

**Table 2.**List of machine independent features that are used to train the dimensionless predictors and how they are derived from the original set of features in Table 1.

Name | Description | Formula |
---|---|---|

${I}_{N}$ | Normalized plasma current | ${I}_{pla}/a{B}_{\varphi}$ |

${f}_{MLA}$ | Mode lock amplitude fraction | $MLA/\left({B}_{\varphi}\right)$ |

${l}_{i}$ | Plasma internal inductance | ${l}_{i}$ |

${f}_{gw}$ | Greenwald density fraction | ${n}_{e}/({I}_{pla}/\pi {a}^{2})$ |

${f}_{P}$ | Radiated power fraction | ${P}_{out}/({P}_{NBI}+{P}_{ICRH}-\dot{{W}_{dia}})$ |

${q}_{95}$ | Edge safety factor | ${q}_{95}$ |

SVM | T2 | RF | GBT | LSTM | |
---|---|---|---|---|---|

Standard | 0.136 | 0.155 | 0.223 | 0.123 | 0.149 |

Dimensionless | 0.189 | 0.294 | 0.291 | 0.123 | 0.283 |

**Table 4.**Train and inference time information for the different models. (* LSTM was trained on a GPU to speed up computation).

SVM | T2 | RF | GBT | LSTM | |
---|---|---|---|---|---|

Train time [s] | 28.21 | 184.89 | 138.52 | 24.88 | 79.18 * |

Inference time [ms] | 0.43 | 0.52 | 0.11 | 0.007 | 0.05 |

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |

© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Croonen, J.; Amaya, J.; Lapenta, G.
Investigation of Machine Learning Techniques for Disruption Prediction Using JET Data. *Plasma* **2023**, *6*, 89-102.
https://doi.org/10.3390/plasma6010008

**AMA Style**

Croonen J, Amaya J, Lapenta G.
Investigation of Machine Learning Techniques for Disruption Prediction Using JET Data. *Plasma*. 2023; 6(1):89-102.
https://doi.org/10.3390/plasma6010008

**Chicago/Turabian Style**

Croonen, Joost, Jorge Amaya, and Giovanni Lapenta.
2023. "Investigation of Machine Learning Techniques for Disruption Prediction Using JET Data" *Plasma* 6, no. 1: 89-102.
https://doi.org/10.3390/plasma6010008