Effect of measurement resolution on data-based models of thermodynamic behaviour of buildings

Abstract

Multiple studies have investigated and shown a theoretical potential in utilising Model Predictive Control (MPC) of residential heating systems to lower CO2-emissions. However, there are several practical issues in realising this potential. This paper reports on a simulation-based study focused on two of these issues both related to the data-based identification of a black-box state-space model for MPC. First, it is investigated how the measurement resolution of the heating energy consumption affects the precision of the model used for MPC. Second, the resolution analysis is combined with an investigation on whether it is possible to obtain appropriate models using data generated from excitation signals that in theory do not lead to occupant discomfort. The performance of the models was evaluated by combining different resolutions of data with different types of excitation signals. The results show that a Pseudo-Random Binary Sequence signal within a temperature span from 20 to 24 °C, and a time and data resolution of one hour and 0.1 kWh, respectively, of the heat consumption is expedient to ensure black-box models sufficient for MPC purposes.