In this article, we are going to highlight the interest in using the data resulting from the future 32 millions of communicating meters which will equip all the French customers from now to 2013, in order to build up Court-terms forecasts and Means-terms concerning the EDF electric consumption (total or by wallets) in an environment ot data stream. First of all we will develop our reflections and afterwards we will evoke some tracks of research. These tracks should enable us to approach some modelling and forecasts by aggregation/disintegration of curves as well as modelling and forecasts on Hilbertian data. Finally we will place side by side these ideas with the Stream-Mining type approaches.