Model structure selection of NARMAX models in the frequency domain - Part II: methodology and applications

This two-part paper aims to present and discuss a novel methodology to perform the model structure selection of NARMAX polynomial models. The strategy called frequency orthogonality proposes a compartmentalized analysis of the candidate regressors, where this initial set is split into groups each studied only at some frequency lines. This is possible due to the application of a proper odd-odd multisine signal. The methodology can be useful in many practical cases where the output spectra is predominantly related to the input. In this second part, the methodology for structure selection is explored. Simulated and real examples are used to confirm the methodology efficiency.