Application of the particle filters for identification of the non-Gaussian systems

Abstract

This paper focuses on application of a particle filter for online identification of non-Gaussian systems. Firstly, the Bayesian inference was described and then the particle filter was defined. The particle filter numerically solves a problem of a recursive Bayesian state estimator. Secondly, the parameters of the linear system and two types of the non-Gaussian systems were estimated by application of particle filter. The first system was the classical linear system. The second system was the linear system with a noise which had a different probability distribution than the Gaussian distribution and the last system was the system with a nonlinearity. Thirdly, the parameters of the non-Gaussian systems were estimated with the gradient based method Levenberg-Marquardt. Finally, the results from the particle filter were compared with the results from the gradient based method Levenberg-Marquardt.