Unsupervised learning of energy signatures to identify the heating system and building type using smart meter data

A high-quality building energy retrofit analysis requires knowledge of building characteristics like the type of installed heating system. This means auditing the building in person or conducting a detailed survey, which is not readily scalable for many buildings. This paper presents a data-driven methodology to identify building characteristics from raw smart meter data sets to allow large scale, high-quality building retrofit analysis. We use the concept of energy signatures, a scatter plot with outside air temperature on the x-axis and electricity consumption on the y-axis, which condenses each building's electricity use into one highly informative graph. Using a Support-Vector Regression model we extract the shape of each signature and cluster them subsequently. Dynamic time warping is used to align the signature shapes of all buildings. In two case studies, consisting of smart meter data sets from 408 and 480 buildings respectively, we show that our clusters correlated well to the heating system type and the building type by comparing to building-level metadata or demographic data.