A performance comparison of sensitivity analysis methods for building energy models
Monte Carlo approach, variance-based sensitivity analysis, regression-based sensitivity analysis, Morris method, comparison
The choice of sensitivity analysis methods for a model often relies on the behavior of model outputs. However, many building energy models are “black-box” functions whose behavior of simulated results is usually unknown or uncertain. This situation raises a question of how to correctly choose a sensitivity analysis method and its settings for building simulation. A performance comparison of nine sensitivity analysis methods has been carried out by means of computational experiments and building energy simulation. A comprehensive test procedure using three benchmark functions and two real-world building energy models was proposed. The degree of complexity was gradually increased by carefully-chosen test problems. Performance of these methods was compared through the ranking of variables’ importance, variables’ sensitivity indices, interaction among variables, and computational cost for each method. Test results show the consistency between the Fourier Amplitude Sensitivity Test (FAST) and the Sobol method. Some evidences found from the tests indicate that performance of other methods was unstable, especially with the non-monotonic test problems.
Tsinghua University Press
Anh-Tuan Nguyen, Sigrid Reiter. A performance comparison of sensitivity analysis methods for building energy models. Build Simul, 2015, 8(6): 651–664.