Data-driven load profile modelling for advanced measurement and verification (M&V) in a fully electrified building

Abstract

The process of decarbonising stock will result in a considerable shift in consumption away from fossil fuels and toward electricity. The growing trend of building electrification necessitates a thorough examination from the standpoint of end-use efficiency and dynamic behaviour in order to fully understand the potential for grid flexibility. The problem of accurately representing dynamic behaviour (e.g. electric load profiles) while retaining simple and easy to use modelling approaches (i.e. supporting a “human in the loop” approach to data-driven methodologies) is a challenging task, especially when operating conditions are very variable. For these reasons, we used an interpretable (regression-based) technique called Time Of Week a Temperature (TOWT) to predict the dynamic electric load profiles before, during, and after the COVID lockdown (for nearly 4 years) of a public office building in Southern Italy, the Procida City Hall. TWOT models perform reasonably well in most conditions, and their application allowed for the detection of changes in energy demand patterns, critical aspects to consider when tuning them, and areas for improvement in algorithmic formulation and data visualisation, which will be the focus of future research.