A methodology for auto-calibrating urban building energy models using surrogate modeling techniques

摘要

Owners of large building portfolios such as university campuses have long relied on building energy models to predict potential energy savings from various efficiency upgrades. Traditional calibration procedures for individual building model are time intensive and require specially trained personnel, making their applications to campuses with hundreds of buildings prohibitive. Recently proposed automatic calibration techniques reduce the manual effort during calibration but require hundreds of thousands of energy simulations which increase their cost. To reduce the computational effort of these methods, this paper proposes a methodology that uses a data-driven approximation technique. Instead of brute-force simulations using detailed engineering models, this study employs statistical surrogate models with an optimization algorithm to estimate properties of unknown building parameters. Results demonstrate that when envelope information is available, this workflow yields sufficiently accurate estimates of hard to observe building characteristics, about 500 times faster than traditional approaches.