• Available Formats
    • Options
    • Availability
    • Priced From ( in USD )

About This Item


Full Description

Mixed mode (MM) buildings are a subset of low-energy buildings that employ both natural and mechanical ventilation, often using manually operable windows for natural ventilation, along with other low-exergy cooling systems such as radiant cooling. This combination of systems has proven diffcult to control in practice, in particular due to the potential for occupants to signi cantly impact building performance. Model predictive control (MPC) and rule extraction are promising methods for optimizing MM building systems in an online setting, and for generating usable control rules that can be implemented in practice. Simulation studies were performed to investigate the impact that occupant actions have on mixed mode buildings, and to improve the performance of natural ventilation controls in mixed mode buildings while accounting for uncertain occupant behavior.

Results show that accounting for occupant behavior in building simulations provides useful insight into the robustness of di erent control strategies with respect to the impact of occupant actions. Two approaches to improving natural ventilation controls are applied to the Research Support Facility (RSF) building, a large MM building in Golden, CO, USA. The rst approach seeks to improve on existing control logic by optimizing setpoints, while the second employs MPC and rule extraction to generate all new control logic. Each approach provides insight into potential aws in existing logic and suggests revisions that lead to improved performance in the presence of occupant behavior. Natural and mechanical ventilation controls changes were implemented in the RSF in the fall of 2013.

Pre- and post-implementation occupant satisfaction surveys show that comfort in the building is una ected by the changes, while measured data shows that energy performance has neither increased nor decreased. In a nal study, rule extraction is applied to optimal control datasets for multiple seasons and locations to develop control rules that approximate optimal controller performance in each locale. Converting state information to state-change information prior to applying rule extraction is shown to improve the performance of extracted rules, and it is shown that rules generated using data for a single season or location do not transfer well to other seasons or locations.