A State Space Approach and Hurst Exponent for Ensemble Predictors

Abstract

In this article we propose a concept of ensemble methods based on deconvolution with state space and MLP neural network approach. Having a few prediction models we treat their results as a multivariate variable with latent components having destructive or constructive impact on prediction. The latent component classification is performed using novel variability measure derived from Hurst exponent. The validity of our concept is presented on the real problem of load forecasting in the Polish power system. © Springer-Verlag Berlin Heidelberg 2013.