A Novel Adaptive Kalman Filter for Markov Jump Linear System with Unknown Measurement Loss Probability

This article studies the state estimation problem of Markov jump linear system (MJLS) subject to unknown measurement loss probability. Compare to the conventional single-mode method, the proposed method can adaptively estimate the state and measurement loss probability for MJLS. Specifically, under system dynamic and measurement loss are independent, the system states and the measurement loss probability are jointly inferred based on variational Bayesian (VB) inference. A simulated fermentation process verifies the effectiveness of the proposed method compared with the existing methods.