Real-time optimal control (RTOC) is considered an efficient tool to improve the energy/cost efficiency of air-conditioning (AC) systems by searching optimal control settings (operation modes and set-points) for local control loops. Extensive computation complexity is a crucial issue for the application of RTOC in AC systems at present due to its nonlinearity, high constraints, and expected global optimums, especially for a large and complex system where more decision variables and interactions among components are involved. An event-driven optimal control (EDOC) strategy was recently proposed to reduce the computation load by conducting optimization actions only when necessary. Different from the traditional time-driven optimal control (TDOC) where the optimization is carried out with a constant frequency, the EDOC triggers optimization actions by events specified as random state transitions. Preliminary studies demonstrated that the EDOC can reduce the computation load of optimization significantly through capturing dynamic changes in the operating conditions and optimizing control settings at a right time.

Considering that the occurrence of events could imply the necessity to execute the optimization, this thesis investigated the necessity of optimization actions, based on which a systematic method to develop EDOC rules was proposed, which can naturally evolve the RTOC from the time-driven to the event-driven paradigm for central AC systems. Conceptually, an optimization action is necessary when the current value of the decision variable is no longer optimal. The variation of the decision variable’s optimal values between two optimization actions was thus used to indicate the optimization necessity. State transitions that have a critical impact on the variation of the decision variable’s optimal values were determined and used to define using the operating data of TDOC for AC systems. The event-action map was then established, which specifies which decision variables to be optimized when a certain event occurs. Multiple evaluation indexes were used to evaluate the performance of the EDOC for AC systems from several aspects including energy use, computation load, and optimization necessity.

The proposed method to develop an EDOC method was applied to three typical central AC systems: a simple AC system, an AC system with multiple chiller plants, and an AC system with multiple AHUs. These applications aim to demonstrate the EDOC developing process and evaluate its performance. Studies of all three systems showed that EDOC strategies could achieve a comparable energy-saving effect compared to the time-driven paradigm and the computation load of the EDOC could be reduced significantly by removing unnecessary optimization actions and decreasing the search dimension of the optimizations.

In the current optimal control of AC systems, the model-based optimal control is in the majority. System/component models are used to predict the energy performance of systems or components, and the model error exists universally and cannot be avoided in practice. Therefore, the impact of the model accuracy on the energy performance of both the EDOC and TDOC methods was investigated in this thesis. Model errors may have a negative impact on the performance of optimization actions, i.e. a certain optimization action may lead to power use increase (negative reward) instead of power use decrease (positive reward). Such an impact was firstly analyzed theoretically to explain how negative reward occurs. Then, different sizes of model errors were investigated using numerical studies, and the possibility of positive/negative reward was quantified in a stochastic way. It was revealed that the event-based optimal control could significantly reduce the possibility of negative rewards when compared with a time-based optimal control, moving a step towards “doing optimization at the right time”.