Dynamic prediction of building HVAC energy consumption by ensemble learning approach

Weisong Tian, Chengwei Lei, Mucun Tian

Abstract: Heating, ventilation, and air conditioning (HVAC) loads contribute a significant portion of energy that a building consumes. Especially for microgrid power systems that operate in island mode, the HVAC loads during daily peak hours can even affect the power quality of the whole grid. Therefore, studying the load behavior of HVAC systems can decrease the energy waste and reduce the risk of overload. However, it is difficult to model the HVAC load due to the essential complexity of the heat transfer mechanisms. In this work, we adopt ensemble learning approaches that consider the load model as a black box and use historical weather data as input and power consumption data as output, to train the HVAC power load model of a building. After setting up the connection between ambient weather conditions and power demands, we are able to dynamically estimate the HVAC power demand for a given day using the weather forecast information.