simplified building models, building energy simulation, complexity management
All research studies based either on analytical and numerical simulations or on testing facilities under controlled or monitored external conditions require some simplifications with respect to the real physical phenomena. The simplification level has to be accurately defined since it introduces some boundaries to the achievable results and to the possibility to generalize the research findings. In particular, this last aspect is very important: once the model complexity is chosen, the desired number of cases can be determined. As a rule, the more complex a model, the larger the sample required to get statistically significant results. In building simulations, we can distinguish two kinds of analyses: the studies aimed to analyze the building performance (e.g., the kind of glazing with the best energy performance in a certain climate, the optimum insulation thickness, the most convenient refurbishment approach) and those focused on building modelling, methods and assumptions (e.g., weather inputs, solar radiation and sky luminance models or algorithms for the calculation of heat transfer through the opaque envelope). As concerns the first group, when the focus is on existing buildings, the sample has to be representative of the building stock possible variability – especially when reference buildings are not available or their characteristics are not completely suitable for inferential statistics. For new building solutions or technologies, instead, proper samples are needed for their evaluation. In the second case, as well, proper building configurations have to be determined and they strongly depend on the researchers’ aims: for instance, the set for the assessment of energy balance models can be different from those for the evaluation of thermal-hygrometric or visual comfort models. In this work, we propose a method to manage the complexity of variables involved in building simulation studies and to identify groups of simplified building models – “shoe-box building models” – or domains of relevant variables suitable to have statistically significant results. The method is applied as an example to the definition of an appropriate set of configurations for the comparison of TRNSYS and EnergyPlus and to the analysis of the discrepancies of six output quantities, i.e., monthly energy needs, hourly peak loads and time of occurrences of hourly peak loads – both heating and cooling. A set of candidate variables describing the building envelope characteristics are studied for each comparison and those more significant are selected for further analyses by means of statistical screening methods (specifically, Spearman’s rank correlation coefficient). Six different short lists of significant buildings variables are then defined for detailed and extensive statistical studies with a full factorial analysis (or equivalent approaches): while in the preliminary study attention is paid in particular to the main effects with a monthly time-discretization of the outputs, the detailed analyses aim at investigating the interactions between the different variables, with shorter time-discretization and focus on the most interesting quantities with respect to the research targets.