Parameter estimation of switching systems

José Ragot (1) , Abdelfettah Hocine (1) , Didier Maquin (1)

1 CRAN - Centre de Recherche en Automatique de Nancy

Since several decades, researchers have been interested in various types of generalized regression models which admit changing parameter values at different time periods. The so-called regime switching models have given a lot of application in the fields of modelling of complex systems, robust identification, detection of behavior change and more generally in process diagnosis. Here, we examine the case where the change in regime cannot be directly observed but may be estimated from observed variables (the input and the output of the process). For that purpose, the well known EM (expectation-maximisation) approach may be applied; to take into account the switches between the regimes, new variables (generally known as hidden or missing) are introduced in order to construct a complete data likelihood function. In this paper, we show (i) how to directly formulate the estimation problem without introducing new variables, (ii) a natural way to solve the obtained equations using hierarchical calculus. An example is given to illustrate how to use the proposed approach.

Mots clés: Switching system; data classification; clustering; multi-model; segmentation; identification; expectation-maximisation

Domaines: Automatique / Robotique

José Ragot, Abdelfettah Hocine, Didier Maquin. Parameter estimation of switching systems. International Conference on Computational Intelligence for Modelling, Control and Automation, CIMCA'2004, Jul 2004, Gold Coast, Australia. pp.CDROM. ⟨hal-00151298⟩