Lumped parameter thermal modelling for UK domestic buildings based on measured operational data
2016-11-21T16:07:47Z (GMT) by
The development and use of thermal models is an integral part of the design process in existing buildings due for refurbishment. Energy predictions for existing buildings are often based on models which assume thermal property values of the building construction elements. However, once built, the actual thermal properties may differ significantly from their estimated values. Possible reasons include thermal bridging, material distortion and moisture content, sub-standard construction on-site and unavailability of construction details. The uncertainties can be reduced if the modelling process can also make use of operational measurements, such as the fuel use and internal temperatures, which have been recorded in the building during operation. To make use of operational data, performance-based models can be used. Performance-based models rely on measured data for the development of the model s architecture and for informing the estimation of the model parameters that would otherwise be based on the modeller s assumptions of the building s characteristics. One solution to the challenge of using performance-based models for existing buildings is to use the Lumped Parameter modelling approach. The Lumped Parameter modelling technique is often used for performance-based modelling of existing buildings due to the moderate knowledge of the building s physical properties required and the limited operational data needed for model training. This thesis investigates the potential of performance-based modelling techniques for existing UK domestic buildings, based on the Lumped Parameter thermal modelling technique, and the use of measured operational data to inform the model structure and parameters. Operational data have been collected in 20 homes as part of the REFIT project, an EPSRC-funded research project on Smart Meters and Smart Homes (REFIT, 2016). This thesis explores 11 houses from the REFIT dataset and, in particular, the temperature, gas and electricity measurements from the participating households, and develops whole-house and sub-system performance-based models using the Lumped Parameter technique. The suitability of simple performance-based Lumped Parameter models in representing typical UK domestic buildings using mainstream operational data such as temperatures and gas consumption measurements is explored. This thesis concludes on the adequacy of the operational data as measured. High correlations (>0.9) between whole-house average indoor temperatures and individual room air temperature measurements prove the use of averages adequate for representing the main rooms of the houses, whereas individual representation of the house s main rooms in use in the same model can prove challenging. A similar result is observed for whole-house radiator representation and the individual radiators. The relationships between the operational data is explored to inform the model structure and to identify collinearity and multi collinearity in the measurements. In terms of whole-house modelling, when using constraints for the parameter values during the model calibration to the measured data the resulting model parameters can be realistic and a good agreement to the measured data can be achieved (on average an RMSE of 1.03 for air temperature). The most significant parameters affecting the mean value of internal air temperatures are the external envelope resistance Re, the non-inertia elements (e.g. windows and doors) resistance, the window area for solar gains, boiler efficiency and the infiltration rate. The indoor air and internal element heat capacitance had the greatest impact on the swing in the internal air temperature (a 75% decrease in the capacitance value resulted in a 190.70% increase in the standard deviation value on average across the 11 houses). The building envelope heat capacitance and the envelope node positioning were the two parameters with the least impact on the model goodness of fit (a 75% decrease in capacitance and a value of 0.9 in envelope node positioning resulted in a 2.57% and 6.68% increase respectively in the RMSE on average across all 11 houses). Finally, the heating system representation using the Lumped Parameter model showed that the whole-house gas consumption data at the meter level, consisting of gas used for space heating as well as other purposes, is inadequate to drive the heating system model. A temperature threshold (e.g. of 1oC) indicating model overprediction can be used to remove the time-stamps of mixed use gas consumption from the model calibration. The heating system model can then be used to quantify gas consumption for space heating and non-space heating uses. In the 11 houses under study, 82.96% of the total gas consumption served for space heating, with 17.04% serving for other non-space heating purposes.