Přejít k hlavnímu obsahu
top

Bibliografie

Journal Article

Marginalized adaptive particle filtering for nonlinear models with unknown time-varying noise parameters

Ökzan E., Šmídl Václav, Saha S., Lundquist C., Gustafsson F.

: Automatica vol.49, 6 (2013), p. 1566-1575

: GAP102/11/0437, GA ČR

: Unknown Noise Statistics, Adaptive Filtering, Marginalized Particle Filter, Bayesian Conjugate prior

: 10.1016/j.automatica.2013.02.046

: http://library.utia.cas.cz/separaty/2013/AS/smidl-0393047.pdf

(eng): Knowledge of noise distribution is typically crucial for good estimation of a non-linear state-space model. However, properties of the noise process are often unknown in the majority of practical applications. Moreover, distribution of the noise may be non-stationary or state dependent, which prevents the use of off-line tuning methods. General estimation methods, such as particle filtering can be used to estimate the noise parameters, however at the price of heavy computational load. In this paper, we present an approach based on marginalized particle filtering where the noise parameters have analytical distribution. Explicit modeling of parameter non-stationarity is avoided and it is replaced by maximum-entropy estimation based on the assumption of slowly varying parameters. Properties of the resulting algorithm are illustrated on both a standard example and a navigation application based on odometry. The latter involves formulas for dead reckoning rotational speeds of two wheels with unknown radii.

: BC