Customer segmentation based on smart meter data analytics: Behavioral similarities with manual categorization for building types

Among the techniques supporting information services that promote energy conservation, clustering has been widely used for customer segmentation. However, the potential of smart meter data analytics based on clustering techniques for small- and medium-sized enterprises to recognize building types, such as offices, retail stores, and restaurants, has not been investigated sufficiently. Estimating building types based on the actual electricity consumption patterns could be a useful approach for energy service providers because manually defined building types can be inaccurate or unavailable. Thus, we evaluated a Zscore-based method and an operating time estimation-based method for clustering by applying them to an open smart meter dataset and comparing the results with manually defined building types. Although the Z-score-based method was more sensitive to drastic demand change, the operating time estimationbased method was more similar to the manual categorization, suggesting that both methods were reasonable. Buildings categorized by energy service providers as 'Insurance', which are usually used as offices with homogenous consumption patterns, tended not to be subdivided. In contrast, building types with diverse consumption patterns, such as 'Restaurant' and 'Entertainment', tended to be subdivided, suggesting the necessity for considering the actual electricity consumption patterns along with the use types.