Authors:Anne de Moliner (EDF R&D) (Speaker) Hervé Cardot (University of Burgundy)Camelia Goga (University of Burgundy)
In the French electricity company EDF, there is a growing need to estimate the mean electricity consumption curves for small geographic areas such as regions, cities or districts. To that aim we use samples of thousands of individual consumption curves recorded by smart meters at a very fine temporal scale and collected according to a sampling design. To address this problem of small area estimation in a functional context, we propose two approaches. The first one consists of adapting unit level models to functional data. In the second approach, one predicts the curve of each non-sampled unit using non-parametric methods, such as random forest or regression trees adapted to functional data, before deriving the estimation of the mean curve of each area.