An alternative method for stochastic systems identification

Abstract

An alternative method is developed for stochastic systems identification in the presence of coloured noise. Central to this method is that the noise covariance vector, which determines the bias in the ordinary least-squares (LS) estimator, is estimated in the way of making use of delayed plant outputs rather than delayed plant inputs. This is very different from the other existing bias-eliminated least-squares (BELS) methods. While achieving estimation unbiasedness, the developed method has algorithmic advantages over the prefiltering based BELS method. Moreover, its performance is comparable to the other BELS methods. Numerical results well correspond to theoretical predictions.