Variational Bayesian IMM-Filter for JMSs With Unknown Noise Covariances

Abstract

In this paper, we derive a modified interacting multiple model filter for jump Markov systems with unknown process and measurement noise covariances. Using the inverse-Wishart distribution as the conjugate prior of noise covariances, the system state together with the noise parameters for each mode are inferred by the variational Bayesian method. The mixing and output estimates are calculated according to the weighted Kullback-Leibler average of mode-conditioned estimates. Simulation results show the effectiveness of the proposed algorithm.