Premium
An Improved Dual Unscented Kalman Filter for State and Parameter Estimation
Author(s) -
Yu Anxi,
Liu Ye,
Zhu Jubo,
Dong Zhen
Publication year - 2016
Publication title -
asian journal of control
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.769
H-Index - 53
eISSN - 1934-6093
pISSN - 1561-8625
DOI - 10.1002/asjc.1229
Subject(s) - kalman filter , control theory (sociology) , unscented transform , observability , invariant extended kalman filter , extended kalman filter , ensemble kalman filter , computer science , filter (signal processing) , fast kalman filter , mathematics , control (management) , artificial intelligence , computer vision
A dual unscented Kalman filter (DUKF) is used to estimate the state and the parameter simultaneously via two parallel unscented Kalman filters. The original DUKF usually has performance degradation as a result of assuming the control inputs of each filter are constant, which usually are disturbance inputs or systematic measurement errors in the control system. An improved dual unscented Kalman filter (IDUKF) with random control inputs and sequential dual estimation structure is derived and applicable to the system in which the parameter is linearly observed and uncorrelated with the state. The accuracy, observability, and computational efficiency of the new filter are discussed. Then, the expansibility of the IDUKF for nonlinear parameter observed substructures is investigated. Finally, two simulation experiments about space target tracking and typical time series filtering are shown. The theoretical analyses and simulation results demonstrate the following. (1) the IDUKF can obtain higher accuracy than the original DUKF and a comparative accuracy with the JUKF (joint unscented Kalman filter) when the state and the parameter are not strongly correlated; (2) the IDUKF has better applicability than the DUKF when the state is correlated with the unknown parameter; (3) when the modeling error is not ignorable, the IDUKF is more robust and more accurate than the JUKF due to lower sensitivity to the modeling error.