A deep learning framework for energy management and optimisation of HVAC systems. Issue 1 (March 2020)

• Record Type: Journal Article
• Authors: Tien, Paige Wenbin; Calautit, John Kaiser; Darkwa, Jo; Wood, Christopher; Wei, Shuangyu; Jay Pantua, Conrad Allan; Xu, Weijie
• Is Part Of: IOP conference series. Volume 463:Issue 1(2020)
• Publication Date: 2020-03
• Subjects: Earth sciences -- Periodicals; Environmental sciences -- Congresses; Environmental sciences -- Periodicals; 550.5
• DOI: 10.1088/1755-1315/463/1/012026
• Languages: English
• ISSNs: 1755-1307

Abstract

Abstract: To enable heating, ventilation and air-conditioning systems to effectively work for the next generation-built environment by reducing unnecessary energy loads while also maintaining satisfactory thermal comfort conditions, this present work introduces a demand-driven deep learning-based framework, which can be integrated with building energy management systems and provide accurate predictions of occupancy activities. The developed framework utilises a deep learning algorithm and an artificial intelligence-powered camera. Tests are performed with new data fed into the framework which enables predictions of typical activities in buildings; walking, standing sitting and napping. Building energy simulation was used with various occupancy profile schedules: two typical static office occupancy profiles, a schedule generated via the deep learning framework and an actual prediction profile. An office space within a case study building was modelled. Initial results showed that the overall occupancy heat gains were up to 30.56% lower when the deep learning generated profile was used; as compared to the static office occupancy profile. This indicated a 0.015 kW decrease in occupancy gains, which also influenced the increase in building heating loads. Analysis indicates the occupancy detection-based framework is a potential solution for the development of effective heating, ventilation and air-conditioning systems. Additionally, the requirement for the deep learning framework to work for multiple occupancy activity detection and recognition was identified.