Improving the predictive power of simplified residential space heating demand models : a field data and model driven study

Large discrepancies have been found in different countries when comparing real energy use in houses to the theoretical energy use calculated using energy performance of buildings (EPB) calculation methods. This prediction gap has become a major concern in the residential building sector, where the information provided on energy performance certificates is often the only energy related basis to support investment decisions in construction and renovation. Additionally, the simplified energy performance calculation methods are also often used for building stock analyses to support policy making by analysing potential savings on a regional or national level. Following these concerns, numerous studies have focussed on behavioural and physical causes of these prediction errors, e.g. on rebound effect and physical temperature take-back. While literature agrees on many findings (e.g. the importance of user behaviour), there is still a debate on what part of the error is due to user behaviour and what part to physical modelling errors. This can partly be explained by the fact that the size of the reported prediction gaps varies depending on the local building tradition, building performance levels and performance assessment framework. This dissertation pursues the investigation on the prediction gap between simplified calculation methods and real energy use in houses, focussing on the space heating demand in single-family houses in Belgium. Building on analyses on field data from inhabited houses, a new simplified calculation approach is developed and used for sensitivity analyses. The first part of the study is data-driven, analysing data from surveys of inhabitants, field-measurements, energy bills and official energy performance calculations. Two datasets are analysed. The first dataset, analysed in Chapter 2, contains over 500 randomly selected high-performance houses. The second dataset, analysed in Chapter 3, was collected in two uniform neighbourhoods, one with old uninsulated houses, the other with new and well insulated houses. The variation in heating profiles at room level and their correlation with building and user related parameters are further analysed in Chapter 4. Large variations in user profiles were found at room level within each dataset, but also between the different datasets and between both neighbourhoods, regarding e.g. heating and ventilation profiles. More lavish heating profiles were found in the higher performance houses, especially in houses with low-temperature central heating systems. With regard to ventilation, variations of technical characteristics even within one neighbourhood, namely of the installed ventilation flow rates, proved more important than the variations in control settings chosen by the user. The regulatory performance assessment method overestimates the ventilation flow rates in old houses without ventilation system and it does not take into account the fact that the windows that are opened the most are mainly those of the often unheated bedrooms. This explains part of the large overestimation of the energy use in old houses made by the energy performance calculation models. However, the user profiles and uncertainties regarding technical properties are not the only causes of prediction errors. The statistical study on the first dataset also revealed the biasing effects of technical, commonly conservative default values used in the official assessment framework and, by consequence, the importance of the assessors, choosing to use default values or more detailed measured or calculated values. The assessors’ work often proves to be more thorough for high performance houses, using fewer default values. This explains in part the larger prediction error in low performance houses. However, the limited sample sizes and the important associations between the building characteristics and performance levels on the one hand and the types of households on the other limit the validity of extrapolating the findings to building stock level. These associations also question whether it is possible to fully discern all building related causes of prediction errors from all user related causes. The data-driven analyses confirmed findings from literature regarding the strong differences in user profiles and measured temperatures in different rooms. To take this into account, a multi-zone model was developed, which however still follows the simplified and efficient quasi-steady-state approach of the EPB-calculation methods. The model is described and analysed in Chapter 5. It is compared to single-zone modelling approaches from Belgium, Germany and the Netherlands and validated in Chapter 6 by simulations on the old case-study neighbourhood considering the real user profiles. The German and Dutch single-zone approaches include correction formulas for taking into account night-time set-back and the fact that not all rooms are heated. Their predicted energy uses lay in the same range as the results from the multi-zone model for the uninsulated houses. However, the fact that single-zone models cannot take into account the location of the heated and unheated rooms nor the fact that the most heated living area is the area with the largest internal heat gains and the lowest window opening hours explains part of their overestimation of the energy use. Furthermore, the single-zone models showed important biases when comparing different renovation measures in further scenario analyses. Most importantly, all single-zone approaches overestimated the relative energy savings associated with loft insulation at least by a factor of two because they do not take the position of the added insulation into account, laying above the colder, unheated bedrooms. Not only can this lead to biased policy making, design or investment choices, but it also explains part of the larger prediction gap identified at lower compared to higher performance levels in Belgium. Compared with single-zone models, building a multi-zone model considerably increases the modelling workload. In spite of the increased prediction accuracy they can offer, their calculation times and this increased workload explain the lack of popularity of multi-zone models in small residential building projects (e.g. a single-family house) and for building stock analyses. In response, Chapter 7 presents a new approach for making multi-zone simulations using mainly limited single-zone inputs. The approach is based on parametric typologies modelled in building information modelling (BIM) software. In collaboration with the research group SmartLab (UGent), a custom BIM-simulation tool was developed, accepting models from e.g. Revit-software to generate single-zone EPB-models as well as multi-zone calculation models. For buildings without 3D BIM-models, predefined parametric multi-zone typologies are fitted in an automated way to the available single-zone data of the specific building, in order to create 3D replacement models. Results from this data-enrichment approach were compared with results based on original BIM-models of three case-study houses. While very good correlations were found between the original models and the replacement models, the findings stressed the importance of selecting an appropriate parametric typology and identified new challenges for improving the fitting procedure. A separate analysis focussed on the use of this parametric typology approach for building stock modelling. Statistical data from the official EPB-database on 15000 houses served as modelling inputs. This analysis proved the large elasticity of the parametric models, allowing building replacement models for very large numbers of different houses, and thus increasing the representativeness of building stock analyses. Combining the approach with the computationally efficient multi-zone calculation model from Chapter 5 allows for more realistic energy modelling, taking into account important parameters such as the real heating profiles, with hardly any increase in workload or calculation time