Use of publicly available data to create a dataset for data-driven urban commercial building energy intensity classification: Model accuracy, interpretation, and implications of an open data framework in Hong Kong

Abstract

Previous studies relied on structured databases containing energy use intensity (EUI) and building features to develop data-driven urban building energy models (D-UBEMs). However, such databases are not available in most cities. In Hong Kong, energy audit is required for commercial buildings, but only EUI is publicly released without any building information. We assessed the adequacy of input features collected and integrated from public sources to develop a precise D-UBEM. We separated Hong Kong commercial building samples into “below-median” and “above-median” EUI categories based on the bimodal distribution of log10EUI and developed a data-driven binary classifier to predict the category. Using publicly available input features, the average classification accuracy is 78.73%. We used the D-UBEM to reveal the association of “above-median” EUI category with input features, and spatially visualize floor areas of Hong Kong commercial buildings belonging to the “above-median” EUI category. In the discussion, we examined the causes of misclassified buildings and highlighted the limitations of our database from public sources in developing accurate and useful D-UBEMs. We suggested an open-data policy framework for releasing building information alongside EUI data and establishing a central database that contains more precise values for building geometries, usage, and thermal performance in Hong Kong.