Abstract
In this paper, the combined parameter and state estimation issues of state-space systems are considered, and the process noises and observation noises are supposed to be coloured noises. By utilising the data filtering technique, we transform the original state-space system into the filtered system for eliminating the interference of the coloured noise in the state equation, and then we derive a filtering-based extended stochastic gradient (F-ESG) algorithm to estimate the system parameters. For estimating the unmeasurable states, we derive a new state estimator by using the preceding parameter estimates to take the place of the unknown system parameters in the Kalman filter. Furthermore, we propose a filtering-based multi-innovation extended stochastic gradient (F-MI-ESG) algorithm to achieve the higher parameter estimation accuracy. Finally, we provide two simulation examples to test and compare the performance of the proposed algorithms. The simulation results indicate that the F-ESG algorithm and the F-MI-ESG algorithm are effective for parameter estimation, and that the F-MI-ESG algorithm is able to achieve more accurate parameter estimates than the F-ESG algorithm.