Find in Library

The increasing penetration of smart meters provides an excellent opportunity to monitor and analyze energy consumption in residential buildings. In this paper, we propose a framework to process the observed profiles of energy consumption to infer the household characteristics in residential buildings. Such characteristics can be used for improving resource allocation and for an efficient energy management that will ultimately contribute to reducing carbon dioxide (CO<inline-formula> <math display="inline"> <semantics> <msub> <mrow></mrow> <mn>2</mn> </msub> </semantics> </math> </inline-formula>) emission. Our approach is based on automated extraction of features from univariate time-series data and development of a model through a variant of the decision trees technique (i.e., ensemble learning mechanism) random forest. We process and analyzed energy consumption data to answer four primitive questions. To evaluate the approach, we performed experiments on publicly available datasets. Our experiments show a precision of 82% and a recall of 81% in inferring household characteristics.