A model reduction technique for nonlinear systems

Nonlinear nondynamic systems which can be modelled by a linear combination of nonlinear functions are considered. An algorithm, based on correlation techniques, is presented for reducing the number of terms in such a model to a fixed but arbitrary number, n. It is shown that when the model is a linear combination of unorthogonal nonlinear functions the algorithm may, but not necessarily, retain n terms which do not result in the optimal n term model, according to a least squared error criterion. A second algorithm is presented for determining the probability of this occurrence, a priori, and thus permits the user to evaluate the usefulness of correlation techniques as a model reduction method. The first algorithm is then used to reduce the number of terms in roll force setup models for a hot steel rolling mill. The second algorithm indicates that the probability of obtaining a suboptimal n term model is very low in this example, even though unorthogonal functions were used. Extensions to dynamic and stochastic systems are also discussed.