Speaker
Dr
Robert Granetz
(MIT)
Description
To address the challenge of disruption prediction, we
have created large disruption warning databases for both
Alcator C-Mod and EAST by compiling values for a number
of proposed disruption-relevant parameters sampled at
many different times throughout all plasma discharges,
disruptive and non-disruptive, during the 2015 campaigns
on the respective machines. The disruption-relevant
parameters include such intuitive quantities as Ip error
[= Ip – Ip (programmed)], radiated power fraction [=
Prad/Pinput], n/nGreenwald, n=1 mode amplitude, as well
as a number of equilibrium parameters derived from EFIT
reconstructions (q95, elongation, etcetera). Examples
of the evolution of these parameters prior to
disruptions on C-Mod and EAST, will be shown.
The disruption warning databases for C-Mod and EAST each
contain parameter values from well over 100,000 time
slices. This allows one to provide quantitative answers
to such questions as: (1) Is parameter “X” (e.g. Ip
error or n/nG or n=1 mode amplitude) correlated with
impending disruptions? If yes, (2) What fraction of
disruptions do not show a correlation (i.e. missed
disruptions)? (3) What is an appropriate trigger level
for each correlated parameter, and how does the number
of ‘false positives’ vary with the trigger level? (4)
What is the typical warning time, and how does the
warning time vary with trigger level? This fundamental
quantitative characterisation of disruption-relevant
parameters is absolutely crucial for developing any
credible real-time disruption warning algorithms.
These databases are also amenable to the application of
advanced ‘machine learning’ techniques to discern more
complicated dependencies on parameters, and the
development of more advanced warning algorithms.
In principle, the disruption-relevant parameters in the
C-Mod and EAST disruption warning databases could be
available in real-time, and their plasma control systems
could implement a disruption prediction algorithm based
on the analysis of these large databases to provide a
warning with sufficient lead time that could be used to
move the plasma to a less unstable state to avoid a
disruption, or to trigger a disruption mitigation
system.
Acknowledgments: This work supported in part by: US DoE
Grants DE-FC02-99ER54512, DE-SC0010720 and DE-SC0010492,
using Alcator C-Mod, a DoE Office of Science User
Facility
| Country or International Organization | US |
|---|---|
| Paper Number | EX/P3-8 |
Primary author
Dr
Robert Granetz
(MIT)
