Sensitivity and Uncertainty in BREDEM-8 Predictions
Ella Quigley, Loughborough University
As part of the requirement for the MSc, a research project had to be carried out, which was completed in September 2010. The details are summarised below.
A year long research project was carried out into some of the assumptions made whilst using the Building Research Establishment Domestic Energy Model, BREDEM-8. The project would carry out a range of sensitivity analyses on the model for one building under different circumstances. An extensive literature review was conducted looking at similar analyses of building models, and realistic data to input into the model. The BREDEM-8 model had been constructed in Microsoft Excel, and so the majority of the project used this software. The programming language Visual Basic for Applications was used to automate the process of performing the various sensitivity analyses. The data obtained was analysed and conclusions were made.
Aims and Objectives
The main objective in carrying out this project was to determine whether the standard tables of data provided for use with BREDEM models gives a realistic estimation of energy use. When information is not known about a building, such as wall U-Value, then this information can be found in a table provided for use with the BREDEM models. It is the validity of using such standard tabulated data that is being questioned.
It was decided that a differential sensitivity analysis, (DSA) and a Monte Carlo analysis, (MCA) would be carried out, to investigate the sensitivity of the model to changes in its input parameters. A post 1919 semi-detached house was selected to test the model. The building was assumed to be unchanged from it initial construction for the first set of analyses; then changes were made to simulate refurbishment, an extension, and changes to the internal floor layout, the analyses were repeated to see if there were any variation in the behaviour of the results. From a total of around 75 input parameters for the building, 27 were initially chosen to vary, and their distributions were defined.
A number of conclusions were drawn from project, it was determined that the model was linear for most parameters, and near linear for the rest.
Around 1000 – 2000 Monte Carlo iterations are required before the normal distribution of the results takes full shape, however as little as 100 simulations are required to make a good prediction of the average result.
The distribution of the results depended upon the building described, the shape of the probability density function was seen to vary with changes to the size of zone 1.
The base case values were found to significantly under predict the mean energy use predicted by the MCAs however is most likely due to assumptions made about the variation in input parameters.
It is recognised that there is not enough data available on real dwelling performance, as so little monitoring is done of the UK stock.
It was found that care should be taking in making assumptions about inaccessible components of a building as these assumptions can have significant influence on the total energy demand, as was found to be the case with the hot water cylinder volume and its level of insulation.
Finally, it is concluded that tabulated data in BREDEM models cannot accurately define the real input values for all the housing stock, buildings are too variable. Instead it is believed that building models should move to incorporate the uncertainties in input parameters, (even small, assumed uncertainties), to give a better prediction by demonstrating a range of results rather than one unique value.
A project was carried out into the validity of using standardised tabulated data as input parameters in a BREDEM-8 house model. A literature review was carried out into the uncertainty in the model‟s input parameters, looking for real building data. The probability distributions of a number of inputs were defined. Differential sensitivity analyses were performed which varied each input parameter one at a time to determine the effect this had on the results. Monte Carlo analyses were also carried out, in which all the inputs were varied randomly at once over a number of simulations. The sensitivity analyses were created using the Programming language Visual Basic For Applications to increase the efficiency and accuracy of each analysis. The main conclusion drawn from the project is that tabulated data designed to represent large portions of the housing stock is not very accurate. Buildings are subject to so much variability that it is impossible to specify one unique value for an input parameter for a given type of building. A much better solution would be to progress towards models that incorporate some level of uncertainty in the inputs, the technological capabilities certainly exist.
A poster summarising the results from the Master’s project.