Identifying whole-building heat loss coefficient from heterogeneous sensor data: An empirical survey of gray and black box approaches

Abstract

Reliable data-driven models that estimate building envelope properties are indispensable for achieving emissions reduction targets. An extensive body of existing research investigates such methods, but benchmarking is limited and it is often unclear whether the approaches are scalable and robust to diverse building properties. Machine learning approaches, which natively handle complex, multivariate datasets, are rarely applied in this domain.This paper benchmarks seven different methods for characterization of the whole-building heat loss coefficient, including traditional gray box and novel black box approaches. To do so, a dataset of 16,000 simulated buildings is created. The models are benchmarked against ground truth, including an assessment of robustness to climate, construction materials, air-infiltration rate and occupant behaviour. We show the deep learning methods outperform other approaches in terms of accuracy and robustness, but that all of the approaches have limitations that restrain their practical usage. Based on this result, we suggest that further research is required to develop reliable and scalable approaches for the characterization of quantitative envelope properties from sensor data. The model code, data creation pipeline and final dataset used for this work are open-sourced so that future work can expand on this study. We encourage the use of our framework to support innovation in this domain.