Key points are not available for this paper at this time.
Abstract The high capabilities of unscented Kalman filter (UKF) for estimating the state variables of a dynamic system have led to their use for parameter estimation as well. In order to use the UKF to estimate the unknown parameters of a dynamic system, the parameters must be assumed to be in the form of virtual state variables. This paper first shows that this assumption causes some serious challenges. Then, trying to solve this problem, a modified UKF algorithm will be presented. Eventually, using the proposed algorithm, the parameters of a power plant turbine-governor system as a typical dynamic system are estimated and the efficacy of the method is investigated. The results show that the proposed method has good performance and is superior to the conventional algorithm. Purpose – This paper proposes a modified UKF algorithm to estimate the parameters of a dynamic system Design/methodology/approach – In this paper, by changing the point of view to system modeling, an improved version of the UKF-based method was presented. In the proposed version of the UKF algorithm, unlike the traditional one, the whole of the measurement signal samples is used as input in each stage of the estimation process. By doing this, throughout the entire simulation time i.e. within the entire time in which the measured signals exist, the unknown parameters are considered constant. Findings – The effectiveness of the proposed method is demonstrated through an illustrative example in parameter estimation of a TGOV1 Turbine-governor system as a case study. The proposed approach overcomes the shortcomings of the conventional method and shows high efficiency. It can be a useful substitute for the conventional UKF method. Originality/value – The proposed method is an evolutionary method whose evolution principles do not random behavior. It is based on Kalman filter rules and relations and enjoys all the advantages of this filter. It looks similar to a smoothing approach whose practical result is to filter out (in the mean sense) estimates with little physical meaning that normally arise when the number of state-variables is increased, that ultimately might lead the filter to diverge.
Doroudi et al. (Thu,) studied this question.