Willi Braun
University of Applied Sciences Bielefeld, Germany
Martin Schroschk
Center for Information Services and High Performance Computing, TU Dresden, Germany
Vitalij Ruge
Siemens AG, Energy Sector, Erlangen, Germany
Andreas Heuermann
University of Applied Sciences Bielefeld, Germany
Bernhard Bachmann
University of Applied Sciences Bielefeld, Germany
Download article
https://doi.org/10.3384/ecp20169159Published in: Proceedings of the American Modelica Conference 2020, Boulder, Colorado, USA, March 23-25, 2020
Linköping Electronic Conference Proceedings 169:17, p. 159-167
Published: 2020-11-03
ISBN: 978-91-7929-900-2
ISSN: 1650-3686 (print), 1650-3740 (online)
Many algorithms related to Modelica-based simulations
heavily rely on the efficient provision of Jacobian matrices.
Besides the accuracy of the derivative information,
the performance of the derivative evaluation is also
of great interest, since it can have a large share in the total
simulation time. In this paper, we propose two complementary
approaches basing on identification of constant
parts and parallelization to accelerate Jacobian evaluation.
Furthermore, the implementations of these techniques
in the open-source Modelica tool OpenModelica
are discussed. The gained speedup in Jacobian evaluation
is demonstrated on benchmark models of the ScalableTestSuite.
Jacobian Evaluation, Symbolic Differentiation,
Derivatives Computation, Coloring, Sparsity, Parallelization,
Modelica, OpenModelica
