Clustering and cluster label prediction for daily electric consumption curves of residential users

In this contribution, we used clustering methods to establish daily consumption patterns for residential electrical consumption data. We also implemented classification and regression methods to predict the daily consumption pattern of the next day. We applied three different criteria – clustering-, classifier-/regressor- and domain knowledge-based – to determine the best clustering method and number of clusters, analyzed in depth the properties of that clustering (particularly the daily consumption patterns), and found that regression performs better than classification in the label prediction task. The investigation was designed to yield insights that reduce the complexity of the residential electricity consumer market and that can be applied in demand response actions such as load shifting.