Data-driven classification of residential energy consumption patterns by means of functional connectivity networks

Understanding energy consumption patterns in the residential sector is of paramount importance for the design of new energy management strategies that are based on innovative information and communication technologies. Smart metering provides considerable opportunities in this respect and allows for the assessment of household characteristics, behaviors and routines that drive household electricity loads. However, the handling of vast quantities of data delivered by smart metering systems requires advanced data analytics technologies and pattern detection algorithms. For the improvement of load forecasting strategies, a good user aggregation and classification is necessary. In the present paper we address this issue and propose a novel technique for data aggregation that was inspired by network science principles. We operate with a dataset of 1 year hourly measured electricity consumption data of 2201 users. The weekly and annual user load profiles are considered separately. Based on the load time series, we construct functional energy consumption networks and extract the minimum spanning tree. Subgroups with similar consumption curves are then objectively identified by means of a community detection algorithm. The proposed methodology is purely data-driven and facilitates an efficient aggregation of users, reduces heterogeneity, eases the study of the relation with environmental factors such as temperature, and is developed to efficiently handle large datasets in an unbiased manner.