Greifswald 2024 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
P: Fachverband Plasmaphysik
P 7: Magnetic Confinement III
P 7.3: Talk
Tuesday, February 27, 2024, 11:45–12:00, ELP 6: HS 3
Machine learning based fast optimization of free parameters in W7-X edge plasma modeling with EMC3-EIRENE — •Y. Luo1,3, S. Xu1, Y. Liang1,3, E. Wang1, J. Cai1, Y. Feng2, D. Deiter3, A. Knieps1, S. Brezinsek1,3, D. Harting1, M. Krychowiak2, D. Gradic2, E. Flom2, F. Henke2, Y. Gao2, R. König2, A. Pandey2, M. Vecsei2, and A. Dinklage2 — 1Forschungszentrum Jülich GmbH, Institut für Energie- und Klimaforschung - Plasmaphysik, 52425 Jülich, Germany — 2Max Planck Institute for Plasma Physics, 17491 Greifswald, Germany — 3Faculty of Mathematics and Natural Science, Heinrich Heine University Düsseldorf, 40225 Düsseldorf, Germany
EMC3-EIRENE is a powerful tool for simulating edge plasma transport, capable of providing insights into transport parameters based on limited local experimental measurements. However, achieving a closer match between simulations and actual experiments often requires extensive scanning of input-free parameters. To address this challenge, we have developed a machine learning model that, by learning from a simulation database, can predict optimal edge cross-field transport coefficients, based on multiple edge measurements. To quantify the performance of the trained model, we calculate mean squared error in the test set, resulting in an error magnitude of 0.024. Moving forward, our plan is to expand the range of learned parameters and significantly enhance the simulation database, thus trying to employ the machine learning technique for directly forecasting plasma information of all EMC3-EIRENE cells based on local experimental measurements.
Keywords: EMC3-EIRENE; Machine learning; Impurity transport; W7-X