Generalized extreme value for smooth component analysis in prediction improvement

Abstract

In this paper we propose a new preprocessing method for smooth component analysis (SmCA). The smoothness measure used in SmCA depends on the signal extreme values directly. We propose the min/max transformation based on the extreme value distribution providing the more realistic and useful signal characteristic in terms of the smoothness. The full methodology is applied as an ensemble method for the energy load prediction improvement. © 2008 Springer-Verlag Berlin Heidelberg.