Improving the operating efficiency of heating, ventilating and air-conditioning (HVAC) systems is important since a small improvement can lead to substantial energy savings, especially for large-scale complex HVAC systems. Real-time optimization (RTO) is an efficient tool to improve the operating efficiency. Almost all traditional RTO methods utilize the time-driven optimization (TDO) mechanism, in which optimization actions are triggered by time in a periodic manner. However, with the increasing complexity of HVAC systems, the traditional TDO mechanism encounters challenges. Since the state variables in complex HVAC systems (e.g. weather and occupancy conditions) are highly stochastic, the TDO mechanism can easily lead to unfavorable optimization actions, e.g. delayed or unnecessary actions. This is because the TDO with a fixed optimization frequency cannot capture the stochastic changes of the state variables promptly. Increasing the optimization frequency is a simple way to improve the performance of TDO, but it will increase the online computational load. As a result, the TDO can hardly achieve a good balance between the optimization performance (e.g. energy efficiency) and the online computational load, which restricts its practical applications.

The limitations associated with the TDO mechanism call for a reformulation of real-time optimization strategies, which motivate us to develop a new optimization mechanism. The mechanism of event-driven optimization (EDO) is therefore proposed and investigated in this study. The key idea of the event-driven optimization is to use “event” rather than “time” to trigger optimization actions. Because it can realize the concept of “taking optimization actions only when necessary”, the EDO mechanism has potential capability for applications in complex HVAC systems to reduce the computational resource utilization while ensure the system performance. To investigate the potential capability systematically, this thesis develops a comprehensive EDO framework and a design procedure for complex HVAC systems.

Firstly, the EDO framework is established, which contains the EDO strategy, optimal control diagram, and fundamental terms associated with the event. An event is formally defined as a set of state transitions. The core of the EDO framework is the {event, policy, action} structure, which works based on the principle that when event occurs, an action is taken based on the policy (which links events with actions). To facilitate the implementations of the EDO, event attributes, types and mathematical representations are also synthesized.

Secondly, to guarantee and improve the optimization performance of the EDO, a design approach is developed according to the EDO framework of {event, policy, action}. Because state transitions (i.e. events) are numerous in a complex HVAC system, a methodology is developed to identify and establish the event space, which includes three steps to address the problems of identifying critical state transitions, event definition, and event space optimization. Both direct and indirect methods are developed for event space establishment. The direct method is constructed based on the system COP (coefficient of performance) deviation. Two indirect methods are developed based on prior-knowledge and data mining, and are thus termed the knowledge-based and data-based methods.

The effectiveness and performances of these methods are demonstrated through the cases studies performed on the simulation platform. Results suggested that when the system dynamics are stochastic and difficult to predict, the EDO strategy is able to adapt to the changing environment because it has a quicker response to environmental changes. Results also show that the EDO strategy can effectively reduce the computational load while not sacrifice the energy performance because unnecessary optimization actions can be avoided.