Effect of Weather and Occupancy Prediction Uncertainties on the Performance of a Model Predictive Controller Applied to a District Heating System

The performance of model predictive control (MPC) depends on the accuracy of the used controller model and prediction of system disturbances. In research on MPC applied to district heating (DH) networks, either perfect or realistic predictions of weather and/or occupancy are used, but the (quantitative) effect of using realistic instead of perfect predictions on the MPC performance has not been studied. This paper addresses this research gap by coupling an MPC to an emulator model of an existing DH network to perform a simulation-based study in which the MPC using imperfect prediction data is compared to an ideal MPC and the currently used rule-based controller (RBC). The results show that the prediction uncertainties negatively impact the MPC performance, but the quantitative effect is limited and MPC still outperforms RBC: in a three-day winter period thermal discomfort is lower for all MPCs with an increased electricity use of maximum 6% compared to RBC and in the two spring periods thermal discomfort is comparable while the use of electrical energy is reduced by 21-32%. Furthermore, the results highlight the importance of a good state observer.