Paper Conference

Proceedings of eSim 2022: 12th Conference of IBPSA-Canada


Data-driven Approaches for Predictive Thermal Modelling in Small and Medium Commercial Buildings

Brent Huchuk, Farid Bahiraei
National Research Council Canada

Abstract: Small and medium buildings comprise 75% of Canada's commercial and industrial building stock. However, small and medium buildings are currently underserved by energy efficiency tools because of their dispersion and relatively low payback potential. In this work, we leveraged the connected thermostat data from 16 small and medium commercial buildings to develop multi-zone data-driven models to be used for predictive control and planning applications. Specifically, we sought to understand which of the candidate data-driven models (i.e., grey and black box models) could most accurately predict the indoor temperatures across a diverse sample of buildings using only typical data collected from a connected thermostat. Our results show that data-driven models, specifically black box models with appropriately selected parameters, had the best predictive performance.
Keywords: Smart thermostats, machine learning, thermal models, small and medium commercial buildings