Scenario-based MPC for energy-efficient building climate control under weather and occupancy uncertainty

Heating, ventilation and air conditioning (HVAC) systems regulate comfort levels in buildings, but also consume a large amount of energy, which makes them an attractive target for efficiency improvements. In this paper, a novel technique called Randomized Model Predictive Control (RMPC) is investigated to improve the control of existing HVAC systems. RMPC uses weather and occupancy predictions to minimize the building's energy consumption. It accounts for the prediction uncertainties by basing its control actions on a given number of sampled uncertainty scenarios. The main advantage of RMPC over existing methods is the absence of a probabilistic disturbance model. This makes the handling of uncertainties straightforward, even if they are non-Gaussian or non-additive. Moreover, the method of removing adverse samples after solving the initial control problem (RMPC-SR) can lead to a further improvement in the control performance, up to a saturation limit. Although theoretical bounds for choosing the sample sizes are available, our simulations show that only a fraction of these numbers is required for a good performance of RMPC and RMPC-SR. The performance of RMPC and RMPC-SR is investigated through extensive simulations on different models, based on empirically collected data. The results demonstrate that both techniques are attractive alternatives to other Model Predictive Control methods, because they show a higher energy saving potential, and are computationally tractable.