Load profiling and customer segmentation for demand-side management

Abstract: The energy transition is accompanied by massive electrification of uses and sectors such as transport. As a result, the pressure on the electricity grid is increasing, and the time to connect to the power system is lengthening. Deploying new infrastructure is a laborious and expensive process but there are alternatives to exploit the flexibility of the power grid. The deployment of smart meters opens the door to many applications related to flexibility on the consumer side, to reduce peak loads that threaten grid capacity. Targeting the right consumers for Demand-Side Management (DSM) is a prerequisite to maximizing the chances of success of such programs. This degree project replicates and adapts the method developed in [14] to segment residential customers. It consists of encoding Daily Load Curves (DLC) using a dictionary of Typical Load Profiles (TLP) and grouping consumers according to the distribution of their TLP. A temporal analysis of the main TLP reveals different consumption behaviors. Customers are segmented into groups that reflect the degree of volatility of their consumption. This enables a classification based on the potential for Energy Efficiency (EE) or Demand Response (D/R) programs. We address the issue of attribute detection using the distribution of TLP of customers. In particular, several classification algorithms are compared to detect TLP characteristic of Electric Vehicle (EV). The obtained load shapes show consumption peaks at night, which may correspond to the charging time of EV. The method is discussed, especially the choice of the number of load profiles to be included in the dictionary of TLP. It proves to be useful to group consumers with similar consumption profiles and opens the door to applications such as individual household consumption forecasting.