Nonlinear Dynamical System Identification Using Unscented Kalman Filter

M. Javvad Ur Rehman, Sarat Chandra Dass, Vijanth Sagayan Asirvadam

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)

Abstract

Kalman Filter is the most suitable choice for linear state space and Gaussian error distribution from decades. In general practical systems are not linear and Gaussian so these assumptions give inconsistent results. System Identification for nonlinear dynamical systems is a difficult task to perform. Usually, Extended Kalman Filter (EKF) is used to deal with non-linearity in which Jacobian method is used for linearizing the system dynamics, But it has been observed that in highly non-linear environment performance of EKF is poor. Unscented Kalman Filter (UKF) is proposed here as a better option because instead of analytical linearization of state space, UKF performs statistical linearization by using sigma point calculated from deterministic samples. Formation of the posterior distribution is based on the propagation of mean and covariance through sigma points.
Original languageEnglish
Title of host publicationProceeding of the 4th International Conference of Fundamental and Applied Sciences 2016
PublisherAIP Publishing
ISBN (Electronic)9780735414518
DOIs
Publication statusPublished - 29 Nov 2016

Publication series

NameAIP Conference Proceedings
Number1
Volume1787
ISSN (Print)0094-243X

Cite this

Ur Rehman, M. J., Dass, S. C., & Asirvadam, V. S. (2016). Nonlinear Dynamical System Identification Using Unscented Kalman Filter. In Proceeding of the 4th International Conference of Fundamental and Applied Sciences 2016 [020003] (AIP Conference Proceedings; Vol. 1787, No. 1). AIP Publishing. https://doi.org/10.1063/1.4968052