Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
P: Fachverband Plasmaphysik
P 26: Poster Session-Plasma Diagnostics
P 26.19: Poster
Donnerstag, 3. März 2016, 16:30–19:00, Empore Lichthof
Acceleration of Bayesian Model Based Data Analysis through Software/Hardware — Humberto Trimino Mora1, Robert Wolf1, •Dirk Timmermann2, Andreas Werner1, and Jakob Svensson1 — 1Max-Planck-Institut für Plasmaphysik, Greifswald, Germany — 2Universität Rostock, Rostock, Germany
Today's leading fusion experiments set new requirements for control systems as well as data analysis to achieve the desired results. Often the performance of state of the art control systems is limited, thus better solutions for the data analysis and the control of complex systems are needed. Typically, control and data analysis use straightforward processing of signals to derive the parameter of interest. Significant improvement can be reached by incorporating knowledge of the system, or the lack of it, into data analysis. Bayesian analysis provides this by empowering the analysis with a rigorous estimation of the uncertainty while introducing previous knowledge with a prior. However, this analysis currently takes long processing times which makes real time analysis and control feedback infeasible. This project attempts to accelerate this analysis towards a real time solution and presents a first proposal solution designed with highly parallelized reconfigurable hardware. The W7-X Dispersion Interferometer diagnostic model was used to implement a single free parameter hardware analysis. The trade-off between arithmetic precision and parallelization area revealed limitations and showed alternate ways to deal with this analysis. The implementation results posed the question of how to deal with the arithmetic error in the forward modeling using this analysis.