A General Framework for Predictors Based on Bounding Techniques and Local Approximation

This paper introduces a general framework for prediction based on nonparametric local estimation and bounding techniques. A set of historic input-output measurements of the system is stored in a database. When a prediction for a given point is required, data from the neighborhood of this point is retrieved and a prediction is formed. These prediction methods return an interval that bounds the considered system output. The width of the obtained interval prediction reflects the amount of information about the system available at the point to be predicted. In addiction, the midpoint of the interval prediction can be used as central estimate. The contribution of the paper is threefold. First, a general framework that covers previous methods proposed in the literature is presented. Second, the general properties of the framework are analyzed. Third, new predictors based on this framework are proposed. Finally, a benchmark example and a comparative study are provided for illustration purposes.