Vision-based detection and prediction of equipment heat gains in commercial office buildings using a deep learning method

Building energy consumption accounts for a large proportion of energy use globally. Previous works have shown that a large amount of energy is wasted in under- or over-utilized spaces since typical building management systems function based on fixed or static operation schedules. While the presence of occupants and how they use equipment contribute to the internal energy demand and affect the thermal environment. Office buildings are likely to have higher cooling demands in the future due to increasing use of equipment, emphasising the need to develop systems which can better understand (and reduce) the impact of internal gains from equipment and adapt to actual requirements. This project aims to develop a deep learning-based approach which enables the detection and recognition of equipment usage and the associated heat emissions in office spaces. Subsequently, the data can be fed into building energy management systems through the formation of equipment heat gain profile; therefore, building energy usage can be effectively managed. Experiments were conducted in typical offices to generate the corresponding heat gain profiles, and then these were used in building simulation software to assess building performance. It was found that the model can perform equipment detection with an accuracy of 89.3%. While maintaining thermal comfort levels, up to 19% annual cooling energy demand reduction can be achieved by the proposed strategy when compared to that for the building managed by a static scheduled heating, ventilation and air-conditioning system, where in the studies, we focus on three types of equipment - computer, printer and kettle that are widely used in the office buildings. The findings indicate that it is feasible to use the deep learning approach to predict equipment heat emission for achieving effective building energy management therefore to reduce building energy demand.