SVRPF: An Improved Particle Filter for a Nonlinear/Non-Gaussian Environment

Abstract

The performance of a particle filter (PF) in nonlinear and non-Gaussian environments is often affected by particle degeneracy and impoverishment problems. In this paper, these two problems are reassessed using the concepts of importance region (IR) selection and particle density (PD), where IR describes the distribution region of particles, and PD describes the density of particles in IR. Based on these two factors, a support vector regression PF (SVRPF) is proposed to overcome the problems from nonlinear and non-Gaussian environments, especially in regard to narrow observation noise. Furthermore, the consistency of the SVRPF and Bayes' filtering is demonstrated. A numerical simulation shows that the performance of the SVRPF is more stable than other filter algorithms. Provided that other conditions are the same, when the observation noise variance is 0.1 and 5, the root-mean-square errors of the SVRPF decrease by 0.5 and 0.03, respectively, compared with that of a general PF.

Wrapped Particle Filtering for Angular Data

“…In the state-of-art particle filtering, the proposal density is approximated as an appropriate Gaussian density [21]- [23] and is accordingly characterized by its mean and its covariance. The proposal density may be defined locally or globally [21], [24] [25]. A globally defined proposal density generates all the particles at each time step [24].…”

“…The choice of proposal density is further affected by: i) whether its shape is close to the unknown posterior PDF and ii) whether any multivariate integrals involved in computing its moments can be computed accurately in real time. The literature primarily uses Gaussian shape for the proposal density, while many contributions appeared in the literature by introducing different numerical approximation techniques [21]- [23], [25]- [28]. Similarly, the literature also witnesses many developments by advancing the resampling techniques [5], [29].…”

Battery remaining useful life prediction using improved mutated particle filter

“…Gilks and Berzuini 26 proposed a particle dilution resampling algorithm based on the Markov chain. As a result, PF has gradually been accepted as a RUL prediction method with its wide applications in nonlinear and non‐Gaussian systems 27‐29 . PF is a Bayesian state estimation algorithm.…”

“…As a result, PF has gradually been accepted as a RUL prediction method with its wide applications in nonlinear and non-Gaussian systems. [27][28][29] PF is a Bayesian state estimation algorithm. It uses some weighted particles to represent the posterior probability distribution of states.…”

An Improved Particle Filtering Technique for Source Localization and Sound Speed Field Inversion in Shallow Water

“…The probability density functions (PDFs) of parameters estimated in a dynamic system are usually non-Gaussian, and the measurement equation is highly non-linear. Although the PF is versatile and robust in handling nonlinear and non-Gaussian estimation problems [21,22], a well-known problem of degeneracy results in a poor estimation performance. The particle degeneracy problem in classical PF seriously limits its development.…”