Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA


Beyond normal: Guidelines on how to identify suitable model input distributions for building performance analysis

Giorgos Petrou 1, Anna Mavrogianni 2, Phil Symonds 2, Mike Davies 2
1 UCL Energy Institute, United Kingdom
2 UCL Institute for Environmental Design and Engineering, United Kingdom

Abstract: Building modelling is a valuable tool in the widespread efforts to decarbonise the built environment. To ensure modelling robustness, uncertainty and sensitivity analysis techniques are often used. Such techniques commonly require model input distributions to be defined. This paper describes a novel approach, within the built environment, for identifying empirically-derived probability distributions of model inputs. Following data cleaning, candidate distributions selected based on measures of skewness and kurtosis are fitted using maximum likelihood estimation. The distribution that best describes the dataset is identified using Akaike Information Criterion and its derivatives along with goodness-of-fit plots. The method was demonstrated for a dataset of wall Uvalue measurements in English homes.
Keywords: Uncertainty, Distribution Fitting, Building Performance Analysis, Building Simulation, Akaike Information Criterion
Pages: 1421 - 1428