Willi Braun
FH Bielefeld University of Applied Science, Germany
Kshitij Kulshreshtha
Universität Paderborn, Germany
Rüdiger Franke
ABB AG, Mannheim, Germany
Andrea Walther
Universität Paderborn, Germany
Bernhard Bachmann
FH Bielefeld University of Applied Science, Germany
Ladda ner artikel
http://dx.doi.org/10.3384/ecp17132363Ingår i: Proceedings of the 12th International Modelica Conference, Prague, Czech Republic, May 15-17, 2017
Linköping Electronic Conference Proceedings 132:40, s. 363-366
Publicerad: 2017-07-04
ISBN: 978-91-7685-575-1
ISSN: 1650-3686 (tryckt), 1650-3740 (online)
Algorithmic differentiation has proven to be an efficient method for evaluating derivative information for implementations of mathematical functions. In the context of the Functional Mockup Interface (FMI) the reverse mode of algorithmic differentiation shows immense promise.
FMI is increasingly used for model-based applications, such as parameter estimation or optimal control. The paper motivates the exploitation of algorithmic differentiation and proposes an extension of FMI for the evaluation of adjoint directional derivatives.
Attempts to interface algorithmic differentiation libraries with Modelica tools have been made. Instead of generating code for the target language which is instrumented with algorithmic differentiation library API and then compiled, in this new approach the intermediate representation used by the library is generated directly. This avoids compilation of the target language that often takes a large fraction of the overall simulation time. It also avoids model execution in order to create such an internal representation at runtime. The initial results are presented here.
Inga referenser tillgängliga
