Master's Thesis Marcus VogtCopyright: EBC
As buildings account for one third of global energy consumption, the reduction in the energy requirements of buildings is important for the overall strategy of energy saving. Dynamic building energy simulations can be used for this, for example, to propose retrofit measures or to calculate future energy savings.
A well-known problem in this matter are the discrepancies between the simulated and measured data, which call for calibration techniques to obtain more accurate building models with more reliable results.
The most recognized calibration techniques use statistical indices to assess and improve the quality of simulation models. To assess the procedure of calibrating building simulation models, a ranking of six tested statistical indices and their combinations is presented. The evaluation of the ranking is done with an automated method, to consider all possible combinations.
The developed method is applied to a use case, a research campus in Germany, for which extensive measured data are available. In this case, dynamic simulations were performed with hourly time-steps. Based on the generated ranking, the suitability of the indices and their combinations for calibrating building models is elaborated. As the method is automated, it allows high numbers of iterations and therefore a more thorough analysis of the available simulated results.
As a result, the following four statistical indices and their combinations should be used for calibration: the “Coefficient of Variation of Root Mean Square Error” (CV(RMSE)), the “Normalized Mean Error” (NME), the “standardized contingency coefficient” and the coefficient of determination (R2). In the evaluation, provided by the automated method, these combinations occupy higher ranking positions than the commonly used combination of CV(RMSE) and the “Mean Bias Error” (MBE).
The best result shows the combination between NME and R2. In addition, it was found that using an evaluation based on multiple criteria is beneficial, because it reveals eventual deterioration of the results.