Thermal performance of UK dwellings: assessment of methods for quantifying whole-dwelling heat loss in occupied homes

The domestic sector accounted for 28% of UK final energy consumption in 2017 (BEIS, 2018a) and 23% of greenhouse gas (GHG) emissions in 2015. Meeting the UK Government’s target of net zero emissions by 2050 (BEIS, 2019) requires improvement of the energy efficiency of domestic buildings through investment in building fabric (BEIS, 2021c). There exists therefore a need for means to quantify dwelling energy efficiency, to both aid in identification of candidates for efficiency improvements and to provide means of measuring the impact of those improvements once implemented; however, a standard method for measuring whole-dwelling energy efficiency has yet to be established in the UK.

This thesis investigates means by which the in-situ fabric thermal performance of a dwelling—as quantified by the Heat Transfer Coefficient (HTC), a metric describing the whole building heat loss rate in Watts per Kelvin indoor-outdoor temperature difference—can be estimated using on-board monitoring data gathered while the dwelling is occupied. Development of such in-use HTC estimation methods is important since: (i) existing approaches for directly measuring a dwelling’s HTC—such as electric co-heating tests—require expensive and intrusive heating experiments in vacated dwellings, a large barrier to application at stock-level; and (ii) existing approaches for predicting the HTC—from design or assumed fabric and construction—frequently overestimate the actual in-situ fabric thermal performance, a phenomenon known as the fabric thermal performance gap. The ability to use non-invasive on-board monitoring data to quantify stock-level fabric thermal performance of UK dwellings would enable a step-change in the way this performance gap is tackled, allowing mass identification of underperforming stock segments, and enabling large-scale evaluation of the impact of remedial works and regulatory instruments.

Four main phases of analysis were conducted, with the ultimate aim of quantifying the fabric thermal performance gap in UK dwellings. In the first phase, application of three in-situ HTC estimation methods to dwelling monitoring data gathered in a semi-detached 1930s-built test dwelling, inhabited by electrically powered simulated occupants, demonstrated their potential suitability for use in occupied dwellings. Underestimation of a reference co-heating HTC by between 18–28% using quasi-steady-state linear regression and averaging methods, and by 3–20% using a dynamic grey-box time series method, illustrated the potential for improved in-use HTC estimation accuracy when using the latter; however, this was at the cost of additional conceptual complexity and computational demands.

Faced with a sparsity of high-quality empirical in-use dwelling monitoring data gathered in dwellings with reliably measured HTCs—as required for robust characterization of in-use HTC estimation accuracies—dynamic thermal simulation was identified as a means of producing hypothetical in-use dwelling monitoring data representing diversity in fabric and environmental conditions in the UK domestic stock. Thus the second phase of analysis sought to establish the suitability of dynamic thermal simulation, using the EnergyPlus simulation engine, for producing simulated monitoring data accurately representing the energy demand and dynamic thermal behaviour of an in-use UK dwelling, specifically the test dwelling analysed in phase 1.

Verification of predicted energy demand and indoor air temperature evolution against empirical data—coupled with consistency between a mean simulated co-heating HTC and the real dwelling’s co-heating HTC—indicated successful model representation. However, further analysis of simulated electric co-heating data revealed EnergyPlus-simulated elemental thermal transmission rates of opaque planar envelope elements were between 10–30% lower than those predicted by U-values calculated using standard methods, indicating potential incompatibility between predicted HTCs and in-situ measurements.The third phase of analysis sought to characterize the impact of building and environmental factors on the accuracy and uncertainty of two in-use HTC estimation methods: a quasi-steady-state averaging method (AVG), and a dynamic grey-box time series method (CTSM). Each method was applied to simulated inuse dwelling monitoring datasets generated for variants of a base case model—closely following after the verified model of phase 2—in which individual building or environmental factors were altered. The factors explored were: monitoring season and length of analysis interval; introduction of unmonitored metabolic gains; perturbation of solar radiation, indoor and outdoor temperatures, electrical gains and metabolic gains; exposure to differing weather conditions and locations; and variation in building fabric. After restricting analysis to a November–February monitoring season due to poor performance at other times, both methods were prone to systematic underestimation of the in-use HTC. Variation in building fabric had the largest impact on the mean degree of underestimation, which appeared to be increased in better-insulated dwellings. The systematic error in both methods, as applied to 30-day-long intervals, was estimated at approximately-20%, with residual uncertainties in mean, bias-adjusted in-use HTCs estimated at -34% to +5% for the AVG method, and -10% to +11% for the CTSM method.

The final phase of analysis assessed the applicability of simulation-derived systematic bias and uncertainty estimates to analysis of real in-use UK dwellings. The AVG and CTSM methods were applied to a newly available in-use dwelling monitoring dataset gathered in 11 dwellings in North West England—the T11 sample—each of whose HTCs had been experimentally measured. Resultant empirically-derived mean bias errors of approximately 0% (AVG) and 4% (CTSM) were inconsistent with simulation-derived bias estimates; furthermore, application of bias adjustments following after simulation results increased the mean degree of error in AVG- and CTSM-derived HTC estimates. The adopted simulation-based approach to characterizing in-use HTC estimation accuracy was concluded to be insufficient and inappropriate for predicting in-use HTC estimation accuracy for dwellings departing from the non-representative simulation sample. Further analysis of larger samples was therefore deferred, pending future work to more robustly characterize in-use HTC estimation accuracy using high-quality empirical data.

While empirically-derived uncertainties of the AVG and CTSM methods obtained from the T11 analysis (±23% and -23% to 32%, respectively) exceeded quoted ±15% uncertainties for the supplied co-heating HTCs, they were low enough to demonstrate potential of the methods as low-cost, non-intrusive alternatives to co-heating. Furthermore, although these uncertainties were large enough to preclude recommendation of the methods for identifying either the fabric thermal performance gap or impact of retrofit in an individual dwelling, per-dwelling mean bias errors were sufficiently small and narrowly distributed to suggest capability to reliably quantify average fabric thermal performance at stock level, with 95% confidence intervals of ±9% (AVG) and -5% to 14% (CTSM) calculated for the T11-wide mean bias error. However, the nonrepresentative nature of the T11 sample prevents generalization to the UK domestic stock.

This research has demonstrated the potential capability of in-use HTC estimation methods as low-cost alternatives to expensive and invasive thermal characterization tests, and produced positive assessment of their future usefulness for quantifying fabric thermal performance at stock level. It has also demonstrated the shortcomings of a solely simulation-based approach to characterizing the accuracy of in-use HTC estimation methods. The results will be of value to building assessors and energy service companies seeking to quantify building fabric thermal performance, policymakers and regulators seeking to improve stock-level dwelling energy efficiency and close the fabric thermal performance gap, and the community of academics and innovators seeking to further develop methods and technologies for assessing in-situ fabric thermal performance using on-board monitoring of occupied dwellings.