A Data Mining Approach to Discover Critical Events for Event-Driven Optimization in Building Air Conditioning Systems

While online optimal control is regarded as an efficient tool to improve the operating efficiency of air conditioning, traditional optimal control strategies utilize the so-called time-driven optimization (TDO) scheme which triggers actions by "time". Although it works well for simple air conditioning systems, several limitations are encountered when systems become more and more complex. Since TDO is a periodic scheme, it may not be suitable or efficient to react to stochastic operational changes. Recently, in order to solve those limitations, the event-driven optimization (EDO) scheme has been proposed, in which actions are triggered by "event". However, previous studies only used prior knowledge to discover important events, which could only find events for general systems, and might not comprehensive because human prior knowledge is limited after all. Moreover, prior-knowledge-based method is able to discover new knowledge. Thus, this paper presents an effective data mining approach to discover the hidden knowledge in massive data set for EDO in building air conditioning systems. Results shown that data-mining-based EDO achieves a higher energy saving with reduced computation load, in comparison with the traditional TDO. Since the data mining approach can help to automate the process of finding critical events and event threshold, it also improves the practicability of EDO.