Data-driven methods for building control — A review and promising future directions. (February 2020)

• Record Type: Journal Article
• Authors: Maddalena, Emilio T.; Lian, Yingzhao; Jones, Colin N.
• Is Part Of: Control engineering practice. Volume 95(2020)
• Publication Date: 2020-02
• Subjects: Heating ventilation and air-conditioning (HVAC) -- Building control -- Model predictive control (MPC) -- Machine learning -- Reinforcement learning; Automatic control -- Periodicals; 629.89
• DOI: 10.1016/j.conengprac.2019.104211
• Languages: English
• ISSNs: 0967-0661

Abstract

Abstract: A review of the heating, ventilation and air-conditioning control problem for buildings is presented with particular emphasis on its distinguishing features. Next, we not only examine how data-driven algorithms have been exploited to tackle the main challenges present in this area, but also point to promising future investigations both from theoretical and from practical viewpoints. Rule based control, reinforcement learning, model predictive control (MPC), and learning MPC techniques are compared on the basis of four attributes that we expect an ideal solution to possess. Finally, on-line learning MPC with guarantees is recognized as an approach with high potential that needs to be further investigated by researchers. Such a solution is likely to be accepted by practitioners since it meets the industry expectations of reduced deployment time and costs. Highlights: The heating, ventilation and air-conditioning (HVAC) control problem is reviewed. Different data-driven algorithms are analyzed in the HVAC context. On-line learning MPC is regarded as a promising solution to be further investigated.