Abstract

We are at the intersection of three major trends in the built environment where: (i) occupants' comfort, health and safety requirements are needed to support a productive workplace while maintaining a low operating cost, (ii) economic and environmental advantages are favouring an increased use of renewable energy generation and to reduce our reliance on fossil fuels, and (iii) major utilities will require regulation and are gradually shifting towards a more dynamic energy market. This thesis contributes a modelling and control framework that unifies and addresses these three points together.

This thesis contributes a methodology for the development of a bootstrapped ensemble-based low-order data-driven grey-box thermal models for supervisory-level optimal controls. The model is integral to a robust sampling-based predictive control (MPC) framework. This approach is directly applicable to most commercial buildings operating on a schedule and can be extended to consider occupant-driven spaces.

The methodology is applied to the Varennes Net-Zero Energy Library: Canada's first institutional net-zero energy building. Exogenous inputs are modelled to consider likely probabilistic outcomes for ambient temperature, cloudiness and interior plug loads. Bounding cases are simulated to contrast the proposed approach against conventional methods. MPC is applied to minimize various cost functions and emphasis is placed on a flexible profile-tracking cost function. The profile to track can be an open-market electrical price or a demand response signal thus improving the grid's flexibility while satisfying the building constraints and better utilizing its systems and storage. In a morning peak demand reduction case, given at least a 4-hour notice, our method is able to pre-heat the building, use minimal energy on-peak and yield the full benefits. Considering a profile tracking case to reduce grid interaction, a 10-12% total energy reduction was achieved for winter where the space was gradually heated in the morning and evening while maximizing HVAC utilization during periods of large photovoltaic generation promoting self-consumption. A similar strategy would be near-impossible to handcraft without optimization-based approaches.

This proposed methodology can guide later implementations in the development of the next generation of low-cost cloud-connected controllers that are easy to deploy and can be adapted dynamically.