Paper Conference

Proceedings of BSO Conference 2018: Fourth Conference of IBPSA-England

     

Utilising Meta-model Based Optimisation to Determine the Effects of Data Granularity on Model Calibration Accuracy

Chris van Dronkelaar, Mark Dowson, Spataru Catalina, Dejan Mumovic

Abstract: Typically, automated calibration is performed using yearly or monthly energy use, masking energy at a higher level of granularity. In this paper, an automated calibration process extended the state of the art, by utilising meta-model based multi-objective optimisation to minimise differences between predicted and measured energy use at a higher level of data granularity. This resulted in a 5-13% monthly average increase of calibration accuracy. It clarified how different convergence criteria in data granularity can mask energy end-uses and can lead to inaccurate calibration results. Several automated calibration limitations were identified; manual calibration for full-scale building models is a necessary pre-processing step; objective convergence is not always guaranteed; the meta-model introduced model error when exporting calibrated parameters to the firstprinciple model; and finally, tight control and understanding of variable input parameters is essential.
Pages: 551 - 559
Paper:
bso2018_5C-3