Robust final prediction error criterion for control oriented models validation using L2 − L1 norm

This paper presents a robust final prediction error criterion (RFPE) for Black-box pseudolinear model structure based on the Huber's ρ-norm. In order to deal with large and numerous atypical data named outliers in the prediction error, the tuning constant in the p-norm is extended. Since the large prediction errors strongly disturb the normal distribution, we adopt the gross error model (GEM) as the corrupted distribution model of these errors. We present a robust final prediction error criterion to the Output Error (OE) models validation. Simulation results are given to get the estimated orders from the new criterion.