| 논문명 |
재실자의 영향을 고려한 데이터 기반 모델 개발 / Development of data-driven models with occupant behavior / 2-4 : 빌딩시뮬레이션 I |
| 저자명 |
안기언(Ahn, Ki-Uhn) ; 김용세(Kim, Yong-Se) ; 김영민(Kim, Young-Min) ; 박철수(Park, Cheol Soo) |
| 주제어 |
데이터 기반 모델 ; 가우시안 프로세스 ; 모델기반 예측 제어 ; 재실자 ; 랜덤 ; Data driven model ; Gaussian Process ; Model Predictive Control ; Occupant ; Randomness |
| 요약2 |
A data-driven model is acknowledged since it requires a few dominant inputs to mimic the dynamics of the building systems. In order to ensure the reliability of model prediction, the inputs should be selected considering the correlation among the training data. In such case, the occupant information is able to be used as a one of inputs for the model. The authors developed the Gaussian Process (GP) model, a data-driven approach, to predict the energy consumption of HVAC systems in existing buildings. The occupant behavior was used for the input data, but its randomness significantly affected the performance of model prediction. The randomness of occupant behavior was investigated using a Normalized Cumulative Periodogram (NCP) based on a random walk hypothesis. A wavelet coherence was used to choose the input among the measured data for the GP model. The developed GP models had good prediction performance; however the training data should be updated in real-time to capture the effect of non-stationary occupant behavior. |
