Robust algorithm for system realization/identification

The author address the problem of determining an accurate mathematical model of appropriate order for a dynamic system, using discrete, time-domain, generally noisy output measurements. The problem typically entails two tasks, namely, the realization task of choosing the model form and order, and the identification task of estimating the parameters in the model after its order has been chosen. Recently, the Eigensystem Realization Algorithm (ERA) was developed for both realization and identification of models of dynamic systems. Separately, Minimum Model Error (MME) estimation has been developed for state estimation of poorly modeled dynamic systems. The ERA method works very well if the measurement noise is low, but loses accuracy as the measurement noise increases. The MME is not specifically a realization/identification technique, but is capable of very accurate state estimation in the presence of both significant model error and measurement noise. In this paper, develop and demonstrate a very robust algorithm for system realization/identification which combines features of both the ERA and MME.