Interval Predictor based on Supporting Hyperplanes

Abstract

A new interval predictor for dynamical systems is presented in this work. The aim is to predict the future output of a dynamical system using a prediction model. This work focuses on predictors that return an interval bound. The interval prediction provides upper and lower bounds of the future system output. Given a set of input-output data of the dynamical system, the interval predictor is obtained using supporting hyperplanes of this set. An inner point of this interval can be used as point prediction. The main goodness of the proposed predictor is to provide a trade off between the width of the interval prediction and the prediction error of the point prediction. A design parameter can be used to balance both objectives. The work proposed a cross-validation methodology to tune this parameter. An example with real data is included to illustrate the proposed interval predictor.