Using an Evidence-Based Approach to Improve Calibration of Non-Domestic Building Digital Twins

In 2020, 24% of UK building emissions came from commercial and public buildings operations. Innovative methods are required to reduce energy demand and increase energy efficiency in this sector. Digital twin (DT) simulations prove valuable for this purpose, enabling the virtual testing of energy efficiency measures before implementing changes. However, the accuracy of DT simulations is only as good as their calibration's. This paper explores the advantages of using highresolution, disaggregated metered and occupancy data to improve DT calibration accuracy. The Andrew Wiles Building at the University of Oxford serves as a case study, with three different DT configurations created. The first and second use as-built and operational input data respectively, both with default occupancy. The third configuration encompasses observed operations and occupancy. Energy simulations run for the year 2022 are compared with the building's metered electricity use. The first two DT configurations result in monthly coefficients of variation of the root mean square error (CVRMSE) and normalized mean bias errors (NMBE) which are above ASHRAE Guideline 14's recommended threshold. In contrast, the third DT configuration falls right under the threshold, with a CVRMSE of 6.4% and NMBE of 5.0%, indicating a significant calibration accuracy improvement and showing the importance of accurate occupancy data in this area. Challenges related to missing data and uncertainties are also discussed.