This paper introduces a new rationale for learning nonlinear dynamical systems. The method makes use of an additional identification dataset, obtained without performing a new experiment on the system under study. The data are generated in an automatical manner, starting from a set of experimentally acquired measurements. In order to leverage the additional generated information, fundamental techniques from the machine learning field known as Semi-Supervised Learning (SSL) are employed and adapted. The problem is then cast as a regularized parametric learning problem. The effectiveness of the proposed approach is assessed on various nonlinear benchmark systems via repeated simulations, comparing the obtained results with a standard regularization method for learning parametric models. (C) 2018, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.

Identification of nonlinear dynamical system with synthetic data: a preliminary investigation

Formentin S.;Previdi F.
2018-01-01

Abstract

This paper introduces a new rationale for learning nonlinear dynamical systems. The method makes use of an additional identification dataset, obtained without performing a new experiment on the system under study. The data are generated in an automatical manner, starting from a set of experimentally acquired measurements. In order to leverage the additional generated information, fundamental techniques from the machine learning field known as Semi-Supervised Learning (SSL) are employed and adapted. The problem is then cast as a regularized parametric learning problem. The effectiveness of the proposed approach is assessed on various nonlinear benchmark systems via repeated simulations, comparing the obtained results with a standard regularization method for learning parametric models. (C) 2018, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
2018
IFAC Symposium on System Identi cation
Regularization; Semi-Supervised Learning; System Identification
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1121600
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 9
  • ???jsp.display-item.citation.isi??? 5
social impact