posted on 2020-10-09, 08:42authored byKostas Mourkos
A considerable amount of literature has focused on the risk of overheating in residential buildings. Overheating has attracted increased attention because it consists a major concern for public health. Overheating is often assessed (as in the case of CIBSE TM52 and TM59 guides) with the aid of Building Performance Simulation (BPS) tools. Nevertheless, the literature shows that it is challenging to predict overheating accurately with these tools. Given that the role of BPS tools is constantly expanding in the area of overheating predictions, it is of vital importance to increase their reliability accordingly.
Modelling overheating reliably requires realistic and accurate input data regarding climate (e.g., air temperature, solar radiation), site context (e.g., exterior noise level, shading from surrounding buildings), building fabric (e.g., thermal insulation, optical properties of glazing), building services (e.g., heat gains from uninsulated pipework, operation of the ventilation system) and occupant behaviour (e.g., windows and blinds operation, heat gains from electrical appliances). All these inputs are associated with a high degree of uncertainty. These inputs also all contribute uncertainty when using BPS tools to predict indoor temperature. Hence, it is important to be able to account for them more consistently and accurately. Therefore, the aim of this research was to improve the reliability of using BPS tools in predicting the overheating risk of dwellings in multi-residential buildings. This was achieved by studying in detail three modern energy-efficient flats located in London; flats that are representative of many high-density developments that have been built in London in recent years.
The analysis that was conducted via dynamic thermal modelling and a Sensitivity/Uncertainty Analysis (SA/UA), employing also monitored data, revealed areas that overheating assessments (such as TM59) need to be further improved. First, in terms of providing guidance on how to handle the thermal interaction between communal spaces and the assessed flat. The comparison between monitored and simulated data (following the above guidance) showed that the Root Mean Square Error (RMSE) with respect to the air temperature in the communal corridors was equal to 1.3°C and 1.1°C in the two corridors that monitored data existed; a discrepancy that was considerably lower than that in the adjacent flats. Second, in terms of examining different infiltration/exfiltration pathways. The way that infiltration/exfiltration is modelled can exert a significant impact on the outcome of the analysis; the respective RMSE for the air temperature varied from 1.7°C to 4.4°C. Third, in terms of specifying more realistic input values for numerous parameters. For example, the replacement of the outdoor air temperature with the monitored supply temperature of the ventilation system closed the gap between predictions and monitored data to a great extent; the RMSE was reduced from 3.8°C to 1.4°C. This finding shows that the common assumption made in thermal modelling that the supply temperature of a mechanical ventilation system is equal to the outdoor temperature, might lead to unrealistic results.
Finally, this research demonstrated how the gap between predictions and reality can be bridged efficiently employing the Bayesian paradigm as implemented in this research including in the calibration not only static (i.e., constant over time) but also dynamic (i.e., varying over time) parameters. The calibration reduced the RMSE from 3.8°C to 0.6°C and from 2.8°C to 0.9°C in the two assessed spaces. Nevertheless, it was shown that it was challenging to predict overheating accurately, even with the aid of the calibrated model. In the first assessed room, both the uncalibrated and calibrated models predicted zero overheated hours in comparison to the monitored 8.5 overheated hours. In the second assessed room, the uncalibrated and calibrated models predicted 1.5 and 18.5 overheated hours respectively, whilst the monitored value of the overheated hours was equal to 10.7. This highlights amongst others, the sensitivity of typical metrics employed in current overheating assessments.
The results of this research support the idea that an overheating assessment should incorporate various sources of uncertainty (e.g., infiltration), providing a range of
values of the desired Building Performance Indicator (BPI) instead of a single value of it. Furthermore, an overheating assessment should consider the employment of less sensitive overheating metrics, such as the Continuously Overheated Intervals (COIs) metric; a metric that when the calibrated model was employed, reduced the discrepancy between monitored and predicted overheated hours from 72.9% to 10.9%.