Artificial Neural Networks for Building Modeling in Model Predictive Controls: Analysis of the Issues Related to Unlocking Energy Flexibility (original) (raw)

2020, TECNICA ITALIANA-Italian Journal of Engineering Science

When used to model buildings in model predictive controls (MPCs), artificial neural networks (ANNs) have the advantage of not requiring a physical model of the building, thus simplifying the development of the MPC. However, if the MPC is intended to operate on the HVAC control system of the building to unlock its energy flexibility, some specific issues associated to the use of ANNs could arise. Thus, in this work we give an insight on the use of ANNs in MPCs, focusing our analysis on some of the most relevant ANN architectures that can be used to predict the thermal demand of a building. Furthermore, the integration of MPCs in real buildings, along with the possibility to unlock energy flexibility by intervening on the comfort band, are discussed. Some issues related to the use of ANNbased MPCs combined with the activation of energy flexibility (e.g., difficulty in training the ANN or in managing flexibility by the MPC) are also investigated.