DdBB: A data-driven framework for building energy use benchmarking

Building energy benchmarking is being widely used to evaluate the extent to which a building consumes energy in excess of its peers. With the energy issues becoming increasingly prominent, many energy benchmarking programs have been developed. These programs are usually based on expert experience, simulation, or statistical methods, in practice, however, they hold certain limitations especially for buildings which have a complex structure and a large number of influencing factors. This paper proposes a novel data-driven framework for building energy use benchmarking called DdBB. Unlike traditional statistical or simulation models, in this work, the core problems of building energy benchmarking are handled by data mining techniques combining building data characteristics, which take sensitivity analysis as a feature selection problem, and building grouping as a clustering problem. DdBB incorporates various data mining approaches and mainly includes four parts: data cleaning and statistical analysis, sensitivity analysis, building energy classification, and model performance evaluation. In addition, CBECS2012 is adopted as experimental and evaluation data, compared with widely used benchmarks provided by Energy Star program, DdBB shows better performance on energy predictive accuracy and robustness.