A reinforcement learning-enabled iterative learning control strategy of air-conditioning systems for building energy saving by shortening the morning start period. (15th March 2023)

• Record Type: Journal Article
• Authors: Dai, Mingkun; Li, Hangxin; Wang, Shengwei
• Is Part Of: Applied energy. Volume 334(2023)
• Publication Date: 2023-03-15
• Subjects: Iterative learning control -- Precooling control -- Reinforcement learning -- Building energy efficiency -- Indoor environment control; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2023.120650
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: A novel iterative learning control strategy is proposed for morning start period. A reinforcement learning method is adopted for the control parameter settings. The control strategy achieves even cooling distribution under limited cooling supply. Daily average precooling time was reduced by 10.9% during typical summer month. Daily precooling energy consumption was reduced from 5.1% to 17.8% Abstract: Air-conditioning systems in commercial buildings are usually switched on in advance to precool the indoor spaces to create an acceptable working environment upon the office hour. However, the central cooling systems often fail to provide enough cooling supply capacity due to the high cooling demand at the morning start period especially in hot seasons. In this situation, the imbalanced cooling distribution in the air-conditioning systems often results in large difference of cooling-down speed among different building zones, so that the precooling time has to be extended, leading to significant energy waste. This study proposes a new iterative learning control strategy to properly manage the cooling distribution (i.e., water valve openings of air-handling units) for achieving uniform cooling (i.e., synchronously reaching the indoor dry-blub temperature setpoint) among building zones during the morning start period. A reinforcement learning method (Q-learning) is adopted for the control parameter setting of the developed iterative learning controller. Validation tests are conducted and results show that the proposed control strategy could reduce the daily precooling time up to 12.1% during typical days in Hong Kong by achieving uniform cooling. The daily energy consumption could be reduced between 5.1% and 17.8% by shortening morning start period, corresponding a weekly electrical energy saving between 1, 376 kWh and 2, 916 kWh in the test building.