MULTIVARIATE IDENTIFICATION: A STUDY OF SEVERAL METHODS
Abstract
Several methods for the identification of non-causal, non - parsimonious dynamic models such as impulse and step response functions are investigated. In particular, ridge regression and general regularization methods are considered, and compared with projection to latent structure or partial least squares (PLS) methods. Some effects of correlated input variables and of autocorrelated noise are presented, and the influence of feedback in the data set is discussed.