Kalman recursions Aggregated Online

In this article, we aim to improve the prediction from experts’ aggregation by using underlying properties of models that provide experts involved in procedure. We restrict ourselves case where perform their predictions fitting state-space data Kalman recursions. Using exponential weights, construct different recursions Aggregated Online (KAO) algorithms compete with best expert or convex combination a more less adaptive way. When are recursions, existing results on literature, taking advantage second-order apply our approach and extend it general adversarial setting modeling errors. these new real-data set electricity consumption show how they can forecast performances compared other exponentially weighted average procedures.