## Estimation problems in jump Markov systems |
Back |

Abstract:Switching dynamic models are frequently used in modern engineering applications. Yet there are modelling issues and several theoretical properties that are still open for investigation. In this talk, we address estimation problems for linear stochastic state-space models whose parameters jump in time among a finite set of values, in accordance with the random evolution of a discrete state. First, we discuss Bayesian state estimation for a model with continuous-time dynamics and sampled measurements. In this model, the discrete state follows the laws of a continuos-time Markov chain, so that parameters may vary between measurements. This provides an accurate description of systems that may change at a rate that is comparable to the rate of observations, and has potential applications e.g. in medicine and bioengineering. However, it makes state estimation a rather complicated task. In a fault detection setting, where one jump to an unknown discrete state occurs at an unknown time, we present an algorithm that yields optimal estimates of the switching time and of the system's state at a low computational cost. Next, we move to discrete-time switching models and focus on the "structure estimation" problem, namely, the estimation of the discrete-state trajectory from the available measurements. We emphasize certain limitations of a purely Bayesian solution and outline alternative estimation strategies. |
Type of Seminar:Public Seminar |

Speaker:Dr. Eugenio Cinquemani Information Engineering , University of Padova | |

Date/Time:Apr 27, 2006 17:15 | |

Location:ETH Zentrum, Physikstr. 3, Building ETL, Room K 25 | |

Contact Person:Prof. Manfred Morari | |

No downloadable files available. | |

Biographical Sketch:Eugenio Cinquemani was born in Bolzano, Italy, on May 30, 1976. He obtained the Laurea degree in 2001 and the Ph.D. degree in 2005 under the supervision of Prof. Giorgio Picci, both at the University of Padova, Italy. During his Ph.D., he worked on estimation problems in stochastic hybrid systems with sampled measurements, and on industrial applications of parameter estimation from sparse tomographic data. He has also been working on wavelet-based Bayesian signal estimation. At present, Eugenio is with the Department of Information Engineering of the University of Padova as a research fellow. His current research interests cover analysis and estimation of stochastic switching systems, with particular focus on the structure estimation and identification problems. |