 |
Universität Augsburg
|
|
|
Universität Augsburg
|
|
Frau Carmen Gräßle
Universität Hamburg
spricht am
Dienstag, 28. Mai 2019
um
16:00 Uhr
im
Raum 1008 (L1)
über das Thema:
| Abstract: |
| The major goal of model order reduction (MOR) using Proper Orthogonal Decomposition (POD) is to replace expensive high-fidelity systems by low-order approximations while ensuring a good approximation quality. POD-MOR finds its applications e.g. in multi-query scenarios like optimal control, where a sequence of PDEs has to be solved many times in order to find a minimum to a given cost functional. The replacement of the large-scale systems by low-order surrogates usually leads to large speed up factors. A crucial challenge within simulation based POD model order reduction is its input dependency. In an offline phase, information data about the underlying dynamical system is generated in a full-order simulation. From these so-called snapshots a POD basis is constructed. If a characteristical property is not captured within the snapshot set, this feature will be missing in the reduced-order model. For this reason, the choice of "good" snapshots is very important in order to achieve "good" reduced-order models. In this talk, we will include adaptivity strategies in the offline phase in order to produce "good" snapshots efficiently. On the one hand, we will consider adaptivity with respect to space and clarify how to combine space-adapted snapshots with POD-MOR. On the other hand, we address the challenge of snapshot location in optimal control, where we use time adaptivity in order to find suitable time instances to sample the snapshots. Numerical examples illustrate the proposed concepts. |
| Hierzu ergeht herzliche Einladung. |
| Prof. Dr. Daniel Peterseim |