Semantic search in household energy consumption segmentation through descriptive characterization

With the widespread adoption of smart metering infrastructures, household energy consumption segmentation is receiving increasing attention. The objective is to transform the large volume of household daily load shapes into representative patterns through clustering methods, with the aim of program targeting and customer engagement. In the literature, there exists a high variation in the number of clusters that different studies have adopted. In order to address the challenge in the trade-off between cluster accuracy and ease of interpretation, in this paper, we introduce a data-driven characterization scheme for resultant clustered load shapes, with the aim of facilitating information retrieval of load shapes with specific semantic attributes. The characterization scheme extracts descriptive features from load shapes to explain their temporal pattern. Using segmentation results on a sample data set from Pecan Street Dataport, we show the feasibility of obtaining the semantic representation of load shapes and performing query analysis by accounting for their similarities. Furthermore, as an application case study, we demonstrated the identification/retrieval of suitable households with specific load types for the adoption of PV-battery system, with average self-sufficiency of 80%.