Prédiction de l'activité humaine afin de réduire la consommation électrique de l'habitat

In this paper we propose a learning mechanism to build the behavior of electrical appliances in a smart home context. The aim of that mechanism is to reduce the consumption of appliances whose effect is delayed (for example the heating) while maximizing the inhabitants comfort. Our approach is based on a predictive model of home humans’ activities. That model is built from local information sensed by the appliances. First we detail our mechanism, then we present its evaluation on the SMACH simulation platform. The accuracy of the model is evaluated on its ability to predict the displacements of a family. Evaluations are performed on several time frames. The strategy that is learned by the electrical appliances is compared with the one proposed by our partner EDF. Preliminary results show that our approach is able to reduce the electrical consumption without significatively lowering the comfort of inhabitants.