A scalable, automatic, and evolutionary algorithm for calibrating urban building energy models

• Scalable EA CMA-ES/HDE for UBEM calibration at building and block scale • Large-scale application of optimisation-based calibration to 200 buildings • Representative user-behaviour in archetype-based UBEMs The national building renovation plan, which has been established under the newly approved Energy Performance of Buildings Directive (EU 2024/1275), can be effectively supported by combining Urban Building Energy Modelling (UBEM) with building archetyping. However, due to inaccessible, unavailable, and inaccurate city-wide building energy consumption data—and the high computational cost of iteratively running large-scale energy models—most UBEM applications remain unvalidated and uncalibrated. This work presents a novel methodology aimed at improving the credibility, transparency, and reliability of national building renovation plans. Specifically, it introduces an automatic calibration procedure for archetype-based UBEMs, using a scalable optimisation algorithm CMA-ES/HDE, which combines the Covariance Matrix Adaptation Evolution Strategy with Hybrid Differential Evolution techniques. Depending on the spatial resolution of the available energy consumption data, the procedure was applied both at the single-building and aggregated multi-building scales within the CitySim environment. The methodology was tested on approximately 200 buildings in the municipality of Monthey (Switzerland). Calibration was achieved by minimising the coefficient of variation of the root-mean-square error, CV(RMSE) , between predicted and observed daily space heating and domestic hot water demands. Using a CV(RMSE) threshold of 25 %, calibration was successful in 43 % of cases at the individual building level; the resulting calibrated parameter ranges were then used to guide the subsequent multi-building calibration. While the methodology does not incorporate the engineering expertise typically applied in manual calibration, it remains a practical and scalable approach for large-scale UBEM calibration, providing valuable insights for future applications.