Simulation-Based Performance Evaluation of Model Predictive Control for Building Energy Systems

The performance of model predictive control can be significantly affected by different choices of controller parameters such as the time intervals for model discretization and control sampling. Due to the lack of a systematic understanding on how these parameters affect the control performance, they are usually selected arbitrarily in practice. In this paper, the combined impacts of selected time intervals for model discretization and control sampling on the performance of model predictive control are comprehensively investigated for the first time through detailed simulations. The simulation results reveal that the time interval for model discretization has a much greater influence on the performance of model predictive control than that for control sampling by affecting the prediction performance, cost saving, and computation time simultaneously. In particular, there are three findings on the time interval for model discretization that are specific to the case studies considered herein. First, it affects the performance of multi-period prediction in a non-linear manner so that it is not always better to have a smaller time interval for model discretization. Second, it has four times greater influence on the resulting cost saving from model predictive control than the time interval for control sampling. Last, it affects the computation time significantly. The increase of the time interval for model discretization from 1 min to 5 min can reduce the computation time by roughly 96%. The proposed simulation-based performance evaluation sheds light on the importance of selecting appropriate time intervals for practical applications of model predictive control to building energy management.