Asymptotic Properties of MOESP-type Methods

Abstract

Abstract Precision of the estimated plant model is often the main interest of system identification. In order to take the predictable part of the noise into account, the noise model is estimated together with the plant model by using ARMAX model. In that case, a model reduction procedure will be required in order to obtain the plant model. On the other hand, the plant model can be estimated directly by using output error (OE) model. In this paper, PI-MOESP method and PO-MOESP method are compared by analysing the signal and noise components of the estimated plant model under the assumption that there are no common poles in the plant and the noise models. The magnitude of the noise component in each method is discussed when the past or future horizon varies and it is shown that there is a possibility that PI-MOESP method gives better performance than PO-MOESP method.