Identification of nonlinear systems with Expectation Maximisation and Particle Filtering
DEI PhD Student
DEI - Aula 3B
15 novembre 2011
The work deals with the parametric estimation of nonlinear systems, a well known problem typical when modelling physical systems for which some parameter is not known a priori (e.g. convective heat transfer coefficients,...). When dealing with such a problem, iterative parameter estimation schemes with suitable filtering methods applied to the system dynamics are the classical approaches. The Extended Kalman Filter (EKF) as well the Unscented Kalman Filter (UKF) are typically proposed in the literature for this task. In this work a different solution has been
investigated. The proposed one couples the Expectation Maximisation (EM) algorithm together with a Monte Carlo technique for the filtering and the smoothing problem. Such a randomised technique, usually refered as to Particle Filter (PF) is introduced and compared to the classical ones. The parameter identification of a generic nonlinear system as well as the Data Compatibility Analisys based on measurements coming from a rotorcraft enforce the applicability of the approach.
Area di ricerca:
Controllo, automazione e misure