Implicit Particle Filtering via a Bank of Nonlinear Kalman Filters

Iman Askari,Mulugeta A. Haile,Xuemin Tu,Huazhen Fang

DOI: https://doi.org/10.1016/j.automatica.2022.110469

2023-06-07

Abstract:The implicit particle filter seeks to mitigate particle degeneracy by identifying particles in the target distribution's high-probability regions. This study is motivated by the need to enhance computational tractability in implementing this approach. We investigate the connection of the particle update step in the implicit particle filter with that of the Kalman filter and then formulate a novel realization of the implicit particle filter based on a bank of nonlinear Kalman filters. This realization is more amenable and efficient computationally.

Systems and Control

What problem does this paper attempt to address?

This paper aims to solve the particle degeneracy problem in Particle Filter (PF). Specifically, the paper explores how to alleviate this problem through the Implicit Particle Filter (IPF) method, especially for the state estimation problem of high - dimensional nonlinear non - Gaussian systems. Particle degeneracy means that during the particle filtering process, the weights of most particles will decrease to almost zero, resulting in a poor approximation effect of the target distribution. Although this problem can be partially solved by increasing the number of particles, as the dimension of the system increases, the required number of particles may increase catastrophically, leading to excessive computational costs.

To improve computational efficiency, the paper proposes a new IPF implementation method, that is, using a set of Nonlinear Kalman Filters (KFs) to approximately implement IPF. This method not only retains the advantage of IPF sampling from the high - probability area of the target distribution, but also improves the computational feasibility. The paper first introduces the concept of Implicit Importance Sampling, which is the basis for understanding IPF and provides a theoretical basis for the connection between IPF and Kalman filters. Then, the paper describes in detail how to use a set of nonlinear Kalman filters to approximately execute the IPF process, with special emphasis on the applications of Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF). Finally, the effectiveness of the proposed KF - IPF algorithm is verified by numerical simulation, and the results show that this method significantly reduces the computational time and the required number of particles while maintaining high estimation accuracy.