Sensitivity analysis methods for building energy models: Comparing computational costs and extractable information

Change log
Menberg, K 
Choudhary, R 

Though sensitivity analysis has been widely applied in the context of building energy models (BEMs), there are few studies that investigate the performance of different sensitivity analysis methods in relation to dynamic, high-order, non-linear behaviour and the level of uncertainty in building energy models. We scrutinise three distinctive sensitivity analysis methods: (a) the computationally efficient Morris method for parameter screening, (b) linear regression analysis (medium computational costs) and (c) Sobol method (high computational costs). It is revealed that the results from Morris method taking the commonly used measure for parameter influence can be unstable, while using the median value yields robust results for evaluations with small sample sizes. For the dominant parameters the results from all three sensitivity analysis methods are in very good agreement. Regarding the evaluation of parameter ranking or the differentiation of influential and negligible parameters, the computationally costly quantitative methods provide the same information for the model in this study as the computational efficient Morris method using the median value. Exploring different methods to investigate higher-order effects and parameter interactions, reveals that correlation of elementary effects and parameter values in Morris method can also provide basic information about parameter interactions.

building energy model, sensitivity analysis, Morris method, Sobol method, regression analysis
Journal Title
Energy and Buildings
Conference Name
Journal ISSN
Volume Title
Engineering and Physical Sciences Research Council (EP/L024454/1)
This study was conducted as part of the ‘Bayesian Building Energy Management (B.bem)’ project funded by the Engineering and Physical Sciences Research Council (EPSRC reference: EP/L024454/1). The financial support by travel grants for Kathrin Menberg and Yeonsook Heo as part of the European Commission Marie Curie BIMautoGEN IRSES grant is gratefully acknowledged.