Analysis of the impact of predictive models on the quality of the model predictive control for an experimental building

To increase energy efficiency of the building sector, many measures have been suggested which often require a predictive model of the building to function. Developing these models is one of the crucial challenges hampering pervasive use of these measures. Therefore, this study aims at assessing the impact of using different predictive models in an energy optimization application for an experimental building. First step in achieving this goal is developing various data-driven models for the investigated building in this study. Afterwards, a framework has been developed in which the performance of predictive models in the optimization strategy namely Model Predictive Control (MPC) could be evaluated. The results reveal that common indicators in the literature do no suffice to score the performance of models used in MPC, but another state of-the-art indicator; multi-step ahead prediction error is more suitable for evaluating predictive models deployed in MPC.