Towards the automated extraction of flexibilities from electricity time series

Several recent and ongoing smart grid projects aim at incorporating more renewable energy sources (RES) into the energy production. Among them, the European MIRABEL project tackles this problem by managing flexibilities on energy demand and supply. Typically, this project assumes that some parts of the energy demand can be shifted when the RES production is sufficient, e.g., the washing machine can be turned on when the wind blows. To express these flexibilities, the project introduces the core-concept of flex-offer. Unfortunately, flex-offer data from the consumers is not yet available. Consequently, in order to test and evaluate the MIRABEL prototype, the flex-offers are extracted from the real world electricity consumption time series. In this work, we investigate, discuss, and experiment several ways to automatically capture flexibility within the electricity time series. Particularly, we show that incorporating domain knowledge, for instance, appliance information or appliance usage frequencies, can improve a lot the outcome of the flex-offer generation and, thus, the MIRABEL project global evaluation.