A cluster selection approach to polynomial NARX identification

Structure selection is the most critical task in nonlinear identification. In the framework of polynomial NARX identification, the concept of cluster can be exploited to devise heuristic techniques for this purpose. The aim of this work is to assess and evaluate the performance of a cluster selection approach to the identification of these models. First the method identifies the relevant clusters and then it performs a refinement identification stage, limiting the model structure to the clusters selected in the first stage. Data obtained on a scaled model of a dam buttress subjected to seismic-like excitations generated by means of a shake table are used to test the method and to compare it with classical NARX identification approaches.