On the Impact of Input Data Uncertainty on the Reliability of Urban Building Energy Models

To prevent climate change from progressing rapidly, greenhouse gas emissions should be reduced drastically. The built environment represents a large share of the energy use and has therefore a great potential in overcoming this challenge, by increasing the energy efficiency and integrating renewable energy sources. It is important that these measures are studied on a district or city level to include the synergy effects that result from the heterogeneity of the existing building stock. Therefore, urban building energy models are emerging, allowing to analyse the current status of the building stock as well as to assess possible future scenarios. Establishing a solid urban building energy model involves three tasks: collecting the required input data, generating and simulating the building energy models and validating the simulations results. An accurate quantification of the energy use in dimensions of time and space is essential for applications such as the optimal design and operation of district energy systems. However, an extensive literature review highlights that collecting the required input data to do so is challenging on district or city level, due to the lack of sufficient (and qualitative) data, resulting in a considerable uncertainty and potentially leading to suboptimal decisions based on urban building energy models. Although substantial efforts to gain insight in the required input data have been done, fundamental knowledge about the combined impact of this multitude of (uncertain) input data on the variation and the variability of the building energy use within districts has not yet been established. Therefore, while focussing on quantifying the energy use of existing Flemish residential districts, the main objective of this work is to quantify the uncertainty on the building energy use within districts as a result of the epistemic and aleatory uncertainty of input data. In this work, it is shown that deterministic approaches, as often applied in literature, do not only underestimate the variability between different dwellings, but also neglect the uncertainty on building level. Therefore, this work proposes a probabilistic approach that consists of three aspects. First, methods to define input parameter variations are developed for the purpose of estimating the current building energy performance indicators within districts based on available data, without intensive on-site data collection. Amongst others, a probabilistic building envelope characterisation method is presented, allowing to allocate the U-values for the ground floor, the external walls, the windows and the roof as well as the window-to-wall ratio based on the building location, building geometry and construction year of every dwelling. The method uses quantile regression to characterise the probability distributions of the five parameters, employs a Gaussian copula to draw correlated samples of these parameters and is trained based on the Flemish energy performance certificates dataset. Second, these methods are used to perform an uncertainty and sensitivity analysis of the building energy use within districts, based on 19 uncertain input parameters and 151,500 building simulations. The uncertainty analysis reveals a significant uncertainty on both the annual building energy use for space heating and domestic hot water (characterised by a coefficient of variation of 54%) and its behaviour over time (characterised by a coefficient of variation of 75%). It is also studied how the uncertainty is impacted by the district size, by the sample size, by a reduced set of uncertain input parameters and by including correlations between the input parameters. The sensitivity analysis indicates that the annual energy use for space heating and domestic hot water is highly correlated to the transmission losses, which are mainly determined by the roof and the external wall. To reduce uncertainty within urban building energy models, these parameters should be collected more carefully. Third, and finally, the use of these methods is illustrated to assess the impact of uncertainty on the building energy demand within districts for two applications. The first application focuses on the optimal control of a district heating network, while the second application studies the renovation potential of an existing district under uncertainty of the current status. It is found that the impact of uncertainty can average out on district level, but is significant on lower levels. Therefore, it is highly recommended to include as well as to reduce uncertainty within urban building energy modelling. In conclusion, this work proposes methods to include uncertainty within urban building energy models and presents the impact of input data uncertainty on the building energy use within existing districts as well as within two applications for the Flemish context.

Publication year:2021

Accessibility:Open