Data mining approach for improving the optimal control of HVAC systems: An event-driven strategy

Abstract

Heating, ventilation and air conditioning (HVAC) systems contribute to a major portion of energy consumption in buildings. Real-time optimal control is an efficient tool to improve the HVAC system efficiency. The formulation of the optimal control strategy is, however, challenging due to the lack of quantitative approaches and unified control format. Many previous studies use domain knowledge or experts’ interpretations to develop the optimal control strategy, which is qualitative, time-consuming and labor-intensive. This study proposes a data-mining-powered event-driven optimal control (EDOC) for improving HVAC operation efficiency. The main contributions are: (1) Provide a standard EDOC format to represent HVAC optimal control strategy; (2) Enable the automatic and quantitative formulation of optimal control strategies with minimal expert involvement. The random forest algorithm is adopted to discover event-driven relationships in the operation data. The effectiveness of the proposed approach is demonstrated through simulations. On average, the formulated EDOC strategy increases the energy saving by 0.9%–4.6% compared with a traditional time-driven optimal control. Meanwhile, events can be customized and can adapt to system renovations. The formulated EDOC is easy to use and can be easily understood by engineers and operators, which can be used to guide the optimal control of building HVAC systems.