Lighted-weighted model predictive control for hybrid ventilation operation based on clusters of neural network models

As one of the most significant factors to determine the success of a hybrid ventilation building, the control strategy for the hybrid ventilation operation attracts increasing attentions from both building designers and researchers these years. Recent advancements of the building control strategy have shown the potential to improve the hybrid ventilation operation. In this paper, we demonstrate the development of an advanced data-driven model predictive control (MPC) algorithm, i.e. a light-weighted three phase NN (neural network) model, for controlling the operation of hybrid ventilation buildings. To develop a robust model predictive control algorithm, firstly, different levels of uncertainties that commonly exist in the real world and building simulation are quantified to efficiently train the central model of the MPC and thoroughly test it in the future. In addition, in the model predictive control establishment process, four candidate mathematical models and ten prediction variables were analyzed during the MPC development to investigate its performance on the prediction as well. The results show that the neural network (NN) achieves the best performance considering both prediction performance and computation time. Six variables including the indoor and outdoor air temperature, relative humidity, office and season index and wind speed were finally chosen. At last, we conduct the validation of the algorithm for hybrid ventilation across four cities in different US climates under uncertainties presented in real practice. The comparison between the MPC and the rule-based control clearly presents that the developed MPC could be better at maintaining the thermal comfort of hybrid ventilation buildings while achieving a comparable amount of energy savings.