A clustering variational Bayesian Kalman filter with heavy-tailed measurement noise

Abstract

In order to solve the problem of unknown measurement noise distribution and variance in the Kalman filtering, the paper proposes a clustering variational Bayesian framework, which includes two parts: (1) a real-time clarifying method is to divide unknown heavy-tailed measurement noise into two Gaussian distributions with different parameters (means and variances), (2) an effective real-time method based Variational Bayesian (VB) is to estimate the parameters of the two Gaussian distributions. Simulations demonstrate that the proposed clustering variational Bayesian Kalman filter outperforms the existing Kalman filters in terms of both estimation accuracy and computational complexity.