Interpretable data-driven methods to automate energy model calibration: Southampton UK Highfield Campus study case

In the journey towards zero emissions, the energy transition is a fundamental aspect of the transformation, and the construction industry is one of the most significant contributors to carbon emissions. Due to the magnitude of investments required to meet decarbonisation objectives and to achieve them within given temporal constraints, we must engage in medium/long-term planning. Consequently, a more in-depth understanding of energy consumption is required to facilitate effective decision-making and reduce uncertainty, which could undermine the effectiveness and credibility of measures. The present project leverages interpretable data-driven methods applied to building energy analytics. The case study is the Highfield Campus of the University of Southampton, where 48 buildings are currently served by a natural gas-fueled Combined Heat and Power (CHP) and District Heating (DH) system. As part of the research, individual energy signature models have been developed for the buildings monitored and the entire district energy system (CHP, DH). The observed data cover the periods before, during and after the COVID-19 pandemic, from 2017 to 2022. Two primary results have been achieved. First, by creating a novel formulation of energy signature regression, it is possible to find the best-fitting models and efficiently determine the actual building balance points (for degree-days consumption normalisation). The second was a counterfactual analysis of energy consumption patterns before, during, and after COVID-19 to detect potential changes. Future research will develop digital twins of building models that can analyse data at varying temporal and spatial resolutions to support the University’s decarbonisation strategy.