Krzysztof Arendt
Center for Energy Informatics, University of Southern Denmark, Denmark
Muhyiddine Jradi
Center for Energy Informatics, University of Southern Denmark, Denmark
Michael Wetter
Lawrence Berkeley National Laboratory, USA
Christian T. Veje
Center for Energy Informatics, University of Southern Denmark, Denmark
Download article
http://dx.doi.org/10.3384/ecp18154121Published in: Proceedings of The American Modelica Conference 2018, October 9-10, Somberg Conference Center, Cambridge MA, USA
Linköping Electronic Conference Proceedings 154:13, p. 121-130
Published: 2019-02-26
ISBN: 978-91-7685-148-7
ISSN: 1650-3686 (print), 1650-3740 (online)
The paper presents an open-source Python tool for parameter estimation in FMI-compliant models, called ModestPy. The tool enables estimation of model parameters using user-defined sequences of methods, which are particularly helpful in non-convex problems. A user can start estimation with a chosen global search method and subsequently refine the estimates with a local search method. Several methods are available already and the tool’s architecture allows for easily adding new ones. The advantages of having a single interface to multiple methods and using them in sequences are highlighted on a case study in which the parameters of a Modelica-based gray-box model of a building zone (nonlinear, multi-output) are estimated using 9 different combinations of methods. The methods are compared in terms of accuracy and computational performance
