A pruning method for the identification of polynomial NARMAX models

A pruning mechanism is developed for the identification of polynomial NARX/NARMAX models, in order to systematically delete redundant terms inserted in the structure selection phase. The proposed approach is capable of overcoming some limitations met by classical regressor deletion criteria, especially when the data available for identification are not adequately exciting or the model family does not include all the correct regressors. The effectiveness of the algorithm is shown by comparison analysis on some simulation examples.