Paper Conference

Proceedings of eSim 2014: 8th Conference of IBPSA-Canada


The Effect of Input Uncertainty in Model-based Predictive Control

H. Burak Gunay, William O’Brien, Ian Beausoleil-Morrison, Brent Huchuk, Michael Palmer, Jim Fletcher, Alexandre Pavlovski

Abstract: The adoption of model-based predictive control (MPC) strategies for building systems, such as auxiliary heating/cooling systems, electric lighting systems, operable window and window shading systems, represents a significant potential to reduce the environment impact of buildings. However, the potential of MPC has been challenged with the limited nature of the information available to the controller to make optimal operational decisions. This paper proposes a selfadaptive MPC method: one in which the control system (1) learns the building physics via a sequential non-linear filter —processing the data coming from a network of sensors/meters; (2) predicts its response using a first-order discrete-time state-space model—processing data coming from forecasts; and, (3) adapts the operation of a heating/cooling system by employing the Newton's method in optimization. The potential of this MPC method is demonstrated through a simulation-based study focused on a hypothetical south-facing office. The performance of the MPC method is contrasted to a classical reactive controller. Its sensitivity to varying resolution and availability of sensors, meters and forecasts is studied. The simulation results indicate that such a method could lead to significant energy savings, even with the limitations/challenges related with the availability and resolution of these inputs. Such a system may also enhance the controllability of the indoor climate within the designated setpoints.
Pages: 845 - 858