A data-driven model for the analysis of energy consumption in buildings

Abstract

Data-driven models are gaining traction in Building Energy Simulation, driven by the increasing role of smart metering and control in buildings. This paper aims to enhance the knowledge in this sector by introducing a practical method to analyse heating consumption. The methodology involves the analysis of hourly total heating demand and outdoor temperature measurements to create and calibrate Energy Signature Curves. Importantly, the building Energy Signature Curve is calibrated independently for each daily hour, resulting in a subset of 24 data-driven models. After calibration, a disaggregation algorithm is proposed to distinguish space heating from domestic hot water usage. The method also evaluates the building’s thermal inertia, examining the correlation between the hourly global energy consumption and the outdoor air temperature moving average. It also presents a methodology for improving the DHW heat consumption model. The methodology is applied to a case study of 51 buildings in Tartu, Estonia, with complete yearly demand measurements from the district heating operator. Thanks to the hourly calibration approach, R2 is 0.05 higher on average than the yearly Energy Signature Curve approach. The difference between estimated and measured annual energy consumption is 8% on average, demonstrating the practicality and effectiveness of the proposed method.