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

This content will become publicly available on July 23, 2020

Title: A real-time machine learning-based disruption predictor in DIII-D

Abstract

A disruption prediction algorithm, named DPRF (Disruption Prediction using Random Forests), has run in real-time in the DIII-D plasma control system (PCS) over more than 900 discharges. DPRF naturally provides a probability mapping associated to its predictions, i.e. the disruptivity signal, now incorporated in the DIII-D PCS. Here, we report on disruption prediction accomplishments in terms of shot-by-shot performances, by simulating alarms on each discharge as in the PCS framework. Due to the optimised performance metric chosen to evaluate DPRF, we find that almost all disruptive discharges are detected in average with few hundred of milliseconds of warning time, but this comes at a high cost of false alarms produced. Performances are not satisfying ITER requirements, where the success rate has to be higher than 95%, but this is not completely unexpected. DPRF is trained on many years of major disruptions occurring during the flattop phase of the plasma current on DIII-D, but without any differentiation by cause. Furthermore, we find that DPRF is affected by a relatively high fraction of false alarms occurring during the first 500 milliseconds from the flattop onset. This subtle effect, more evident on the discharges where DPRF ran in real-time, can be marginalized bymore » taking specific precautions on the validity range of the predictions, and performances do improve. Even if presently burdened by some limitations, DPRF provides an incredible and novel advantage. Thanks to the feature contribution analysis (e.g., the identification of which signals contributed to triggering an alarm), it is possible to interpret and explain DPRF predictions. It is the first time that such interpretability features is exploited by a disruption predictor: by uncovering the disruption events' causes, a better understanding of disruption dynamics is achieved, and a clear path toward the design of disruption avoidance strategies can be provided.« less

Authors:
ORCiD logo [1]; ORCiD logo [2]; ORCiD logo [3]; ORCiD logo [2]; ORCiD logo [2]
  1. General Atomics, San Diego, CA (United States); Massachusetts Inst. of Technology (MIT), Cambridge, MA (United States)
  2. Massachusetts Inst. of Technology (MIT), Cambridge, MA (United States)
  3. Princeton Plasma Physics Lab. (PPPL), Princeton, NJ (United States)
Publication Date:
Research Org.:
General Atomics, San Diego, CA (United States); Massachusetts Inst. of Technology (MIT), Cambridge, MA (United States); Princeton Univ., NJ (United States)
Sponsoring Org.:
USDOE Office of Science (SC), Fusion Energy Sciences (FES) (SC-24)
OSTI Identifier:
1562300
Report Number(s):
DOE-GA-54698
Journal ID: ISSN 0029-5515
Grant/Contract Number:  
FC02-04ER54698; SC0014264; AC02-09CH11466
Resource Type:
Accepted Manuscript
Journal Name:
Nuclear Fusion
Additional Journal Information:
Journal Volume: 59; Journal Issue: 9; Journal ID: ISSN 0029-5515
Publisher:
IOP Science
Country of Publication:
United States
Language:
English
Subject:
70 PLASMA PHYSICS AND FUSION TECHNOLOGY; disruption prediction; machine learning; random forests; DIII-D; real-time

Citation Formats

Rea, C., Montes, K. J., Erickson, K. G., Granetz, R. S., and Tinguely, R. A. A real-time machine learning-based disruption predictor in DIII-D. United States: N. p., 2019. Web. doi:10.1088/1741-4326/ab28bf.
Rea, C., Montes, K. J., Erickson, K. G., Granetz, R. S., & Tinguely, R. A. A real-time machine learning-based disruption predictor in DIII-D. United States. doi:10.1088/1741-4326/ab28bf.
Rea, C., Montes, K. J., Erickson, K. G., Granetz, R. S., and Tinguely, R. A. Tue . "A real-time machine learning-based disruption predictor in DIII-D". United States. doi:10.1088/1741-4326/ab28bf.
@article{osti_1562300,
title = {A real-time machine learning-based disruption predictor in DIII-D},
author = {Rea, C. and Montes, K. J. and Erickson, K. G. and Granetz, R. S. and Tinguely, R. A.},
abstractNote = {A disruption prediction algorithm, named DPRF (Disruption Prediction using Random Forests), has run in real-time in the DIII-D plasma control system (PCS) over more than 900 discharges. DPRF naturally provides a probability mapping associated to its predictions, i.e. the disruptivity signal, now incorporated in the DIII-D PCS. Here, we report on disruption prediction accomplishments in terms of shot-by-shot performances, by simulating alarms on each discharge as in the PCS framework. Due to the optimised performance metric chosen to evaluate DPRF, we find that almost all disruptive discharges are detected in average with few hundred of milliseconds of warning time, but this comes at a high cost of false alarms produced. Performances are not satisfying ITER requirements, where the success rate has to be higher than 95%, but this is not completely unexpected. DPRF is trained on many years of major disruptions occurring during the flattop phase of the plasma current on DIII-D, but without any differentiation by cause. Furthermore, we find that DPRF is affected by a relatively high fraction of false alarms occurring during the first 500 milliseconds from the flattop onset. This subtle effect, more evident on the discharges where DPRF ran in real-time, can be marginalized by taking specific precautions on the validity range of the predictions, and performances do improve. Even if presently burdened by some limitations, DPRF provides an incredible and novel advantage. Thanks to the feature contribution analysis (e.g., the identification of which signals contributed to triggering an alarm), it is possible to interpret and explain DPRF predictions. It is the first time that such interpretability features is exploited by a disruption predictor: by uncovering the disruption events' causes, a better understanding of disruption dynamics is achieved, and a clear path toward the design of disruption avoidance strategies can be provided.},
doi = {10.1088/1741-4326/ab28bf},
journal = {Nuclear Fusion},
number = 9,
volume = 59,
place = {United States},
year = {2019},
month = {7}
}

Journal Article:
Free Publicly Available Full Text
This content will become publicly available on July 23, 2020
Publisher's Version of Record

Citation Metrics:
Cited by: 2 works
Citation information provided by
Web of Science

Save / Share:

Works referenced in this record:

Random Forests
journal, January 2001