Assessing the impact of the dataset’s size and quality on building thermal models for energy flexibility

Abstract

The demand side is required to increase its energy flexibility to tackle the demand–supply mismatch caused by integration of renewable energy sources into the energy mix. Model Predictive Control (MPC) is one of the promising measures for achieving this goal. MPC requires operational data from the building to train its predictive model. The quality of this data affects the performance of MPC. This paper aims to analyze the impact of a dataset’s size and excitation method on flexibility harnessed by MPC. Five datasets have been generated that differ in their size and excitation method. They are used to train three modeling algorithms on two buildings. Next, generated predictive models are integrated into an MPC framework, to assess the impact of datasets on the performance of MPCs for activating the energy flexibility of a building. The case studies are two similar dwellings where one is equipped with Electric Heaters (EH), and the other uses an UnderFloor Heating (UFH) system. Results show that the flexibility of the EH case is more sensitive to the size of the dataset. Furthermore, it is shown that the suitability of the excitation signal for flexibility activation depends on the modeling algorithm. The results show that choosing an ill-suited dataset might cost the MPC 35 % and 20 % of the potential flexibility for EH and UFH cases, respectively.