| 논문명 |
강화학습을 이용한 건물 에너지 최적 제어 / Optimal Control of Building Systems based on Reinforcement Learning / 춘계-05. 건축환경및설비 |
| 저자명 |
안기언(Ahn, Ki Uhn) ; 박철수(Park, Cheol Soo) ; 여명석(Yeo, Myoung-Souk) |
| 수록사항 |
대한건축학회 학술발표대회 논문집, Vol.38 No.1 (2018-04) |
| 주제어 |
최적제어 ; 모델 프리 제어 ; 강화학습 ; 딥 큐 러닝 ; Optimal Control ; Model-free Control ; Reinforcement Learning ; deep Q-learning |
| 요약2 |
In the traditional model predictive control (MPC), it is difficult to ensure the reliability of an optimal solution from the simulation model due to the complexity of system dynamics, computational cost, and the uncertainty of the model and the reality. In contrast to the MPC, reinforcement learning learns what to do and how to map the states to actions so as to achieve a goal. This study introduces the on-line optimal control based on the reinforcement learning for the entire of building energy systems including four air handling units, two electric chillers and a cooling tower installed in an office building. In particular, the deep Q-learning, which is one of the reinforcement learnings, is used to control the systems with a model-free approach. This paper presents a case study of deep Q-learning applied to an office building. |
