Abstract
Due to the over-parameterized models in detailed thermal simulation programs, modellers undertaking validation or calibration studies, where the model output is compared against field measurements, face difficulties in determining those parameters which are primarily responsible for observed differences. Where sensitivity studies are undertaken, the Morris method is commonly applied to identify the most influential parameters. They are often accompanied by uncertainty analysis using Monte Carlo simulations to generate confidence bounds around the predictions. This paper sets out a more rigorous approach to sensitivity analysis (SA) based on a global SA method with three stages: factor screening, factor prioritizing and fixing, and factor mapping. The method is applied to a detailed empirical validation data set obtained within IEA ECB Annex 58, with the focus of the study on the airflow network, a simulation program sub-model which is subject to large uncertainties in its inputs.
Highlights
Current detailed building energy simulation (BES) tools are capable of representing the main phenomena determining the thermal and energy performance of buildings
After parameter screening the identified most important factors (MIF) have been grouped according to parameter typology and all the least important factors (LIF) have been lumped in the LIF group
The explained methodology is apt as a preparatory phase to model calibration and was applied in three phases: Factor screening (FS), Factor prioritizing (FP) and factor fixing (FF), and Factor mapping (FM)
Summary
Current detailed building energy simulation (BES) tools are capable of representing the main phenomena determining the thermal and energy performance of buildings. SA is used to aid model validation or calibration In these problems, SA can be used to reduce model dimensionality through parameter screening, and calculate prediction uncertainties in order to better compare simulation outcomes with target measurements. BES produce time series as outputs and in most approaches the calculation of SA indexes is carried out for each time step or by considering integrals of output variables and distances from reference values In the former case, while it is possible to see how the model sensitivity changes over time a large and redundant amount of information is produced which is hard to analyse and summarize. The first part of the paper describes the experiment and the model used as the example application This is followed by details of the UA performed in order to assess prior parameter uncertainties, and a description of the sensitivity methods used. The main results are discussed and the conclusions drawn regarding the efficacy of the methods investigated
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
