Bounding Approach to Parameter Estimation without Priori Knowledge on Model Structure Error

Abstract

Parameter estimation of an autoregressive movmg average (ARMA) model is discussed in this paper by using bounding approach. Bounds on the model structure error are assumed unknown, or known but conservative. To reduce this conservatism, a point-parametric model concept is proposed, where there exist a set of model parameters and structure error corresponding to each input. Feasible parameter sets are defined for point-parametric model. Bounded values on the model parameters and structure error can then be computed jointly by tightening the feasible set using observations under deliberately designed input excitations. Finally, a constantly bounded parameter model is established, which can be used for robust control.