Interpretable data-driven building load profiles modelling for Measurement and Verification 2.0

Accelerating the decarbonisation of the built environment necessitates increasing electrification of end-uses, which in turn poses the issue of rethinking the role of energy efficiency in conjunction with flexibility in grid interaction. This requires a better understanding of the electricity load profiles at hourly or sub-hourly intervals using techniques that are simple, reliable, and interpretable. To this extent, this study proposes a reformulation of the Time Of Week and Temperature modelling approach. This approach is able to separate the energy consumption dependence on building operational characteristics (Time Of Week) and on weather (outdoor air temperature), through a highly automated modelling workflow, necessitating minimal effort for model tuning. These features, along with its intrinsic interpretability due to its formulation using multivariate regression and the availability of open-source software, makes it an ideal starting point for applied research. The case study selected for the research is a fully electrified public building in Southern Italy. The building has been monitored for 5 years, before, during and after the COVID-19 lockdown. The novel model formulation is calibrated using hourly interval data with a Coefficient of Variation of Root Mean Square Error in the range of 20.0–28.5% throughout the various monitoring periods. The counterfactual analysis of electricity consumption indicates a 10.7–26.7% decrease in electricity consumption due to operational adjustments following COVID-19 lockdown, highlighting the impact of behavioural change. Finally, the possibility of additional workflow automation and enhanced interpretability is discussed.