Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA


Reducing the energy performance gap on building stock level using actual energy data

Paula van den Brom, Laure Itard
Delft University of Technology, The Netherlands

Abstract: Building energy simulation models are widely used to predict the efficiency of energy saving measures at building stock level. Previous research has shown that there is a significant gap between actual energy consumption, and simulated energy consumption in buildings and in the building stock. The gap cannot come from differences between assumed and actual occupant behaviour l because occupant behaviour can be expected to be averaged at building stock level, indicating a more structural problem. Because it is important to predict building stock energy consumption accurately this research investigates the use of actual energy consumption data and automatic calibration techniques to improve standard assumptions in building energy simulation models used to assess the whole building stock. A steady state model used in NL in the framework of the EPBD has been tested using particle swarm optimisation. The particle swarm method is selected because it requires relatively few iterations, which means the method is relatively fast. The method was able to reduce the root mean square error of the energy performance gap by nearly 24% and, most important, the average energy performance gap in the sample (133 dwellings) as well as in the control group (180), disappeared almost completely. This method has the potential to make building simulation models a more reliable tool for policymakers.
Keywords: energy performance gap, calibration, machine learning, building stock
Pages: 2070 - 2077