SMuK 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
P: Fachverband Plasmaphysik
P 12: Poster II
P 12.29: Poster
Mittwoch, 22. März 2023, 17:30–19:00, HSZ EG
Machine Learning Applications in Control at ASDEX Upgrade — •Johannes Illerhaus1,2, Wolfgang Treutterer1, Alexander Bock1, Rainer Fischer1, Paul Heinrich1, Frank Jenko1,2, Ondrej Kudlacek1, Gergely Papp1, Tobias Peherstorfer1,4, Bernhard Sieglin1, Udo von Toussaint1,5, Hartmut Zohm1,3, and the ASDEX Upgrade Team6 — 1Max-Planck-Institut für Plasmaphysik, Garching, Germany — 2Technische Universität München, Garching, Germany — 3Ludwig Maximilian Universität, Munich, Germany — 4Technische Universität Wien, Vienna, Austria — 5Technische Universität Graz, Graz, Austria — 6see the author list of U. Stroth et al. 2022 Nucl. Fusion 62 042006
Plasma control is essential for the operation of fusion devices. The individual control tasks depend on high-dimensional and possibly noisy input data and typically have a latency requirement of milliseconds to be real-time capable. Machine learning (ML) models are well suited for this application. While they are often computationally expensive to train, they generally have a cheap, low-latency inference process. Additionally, deep learning models have been shown to be capable of extracting complex hidden interactions in high-dimensional, noisy data. This contribution will illustrate two ML applications in plasma control: real time capable approximations of high-fidelity offline models for kinetic profiles, and deep-learning-based augmentations to the accuracy of the pellet fragment analysis used in the development of the shattered pellet injection disruption mitigation system tested on ASDEX Upgrade for use in ITER.