Towards a Verified ODE Solver for GPU-Based Parameter Identification

Aus Kompetenzportal
Wechseln zu: Navigation, Suche
Towards a Verified ODE Solver for GPU-Based Parameter Identification
Autor Ekaterina Auer, Stefan Kiel, Andreas Rauh
In: Safety, Reliability, Risk, Resilience and Sustainability of Structures and Infrastructure

Proceedings of the 12th International Conference on Structural Safety and Reliability

C. Bucher, B.R. Ellingwood, D.M. Frangopol (eds.)

Ausgabe
ISBN/ISSN: 978-3-903024-28-1
Erscheinungsjahr 2017
Jahrgang
Seitenzahl 2039--2049
Hyperlink
Review peer

In this paper, we implement a simple verified ODE solver to run on the GPU as a first step towards GPU-based fully verified parameter identification. Our goal is to use the proposed parallelization to speed up parameter identification for dynamic SOFC models based on ordinary differential equations. We illustrate the potential of the approach using a simple close-to-life example, for which we compute the cost function under parameter uncertainty on the GPU and compare the performance to the non-parallelized case.