New results on the identification of interval predictor models

In this paper, the problem of identifying a predictor model for an unknown system is studied. Instead of standard models returning a prediction value as output, we consider models returning prediction intervals. Identification is performed according to some optimality criteria, and, thanks to this approach, we are able to provide, independently of the data generation mechanism, an exact evaluation of the reliability (i.e. the probability of containing the actual true system output value) of the prediction intervals returned by the identified models. This is in contrast to standard identification where strong assumptions on the system generating data are usually required