VARIABLE SELECTION IN NON-LINEAR SYSTEMS MODELLING

A new algorithm which preselects variables in non-linear system models is introduced by converting the problem into a variable selection procedure for a set of linearised models. Because on this result an algorithm which consists of a cluster analysis linearisation sub-region division procedure, a linear subset selection routine using an all possible regression algorithm and a genetic algorithm is developed. This algorithm can be applied to the modelling of non-linear systems using a wide class of model forms including the non-linear polynomial model, the non-linear rational model, artificial neural networks and others. Numerical simulations are included to demonstrate the efficiency of the new algorithm.