| Title |
Short-Term Electricity Consumption Prediction based on Occupancy Information Using Deep-Learning Network Models |
| Authors |
Byung-Ki Jeon ; Eui-Jong Kim ; Kyung-Ho Lee ; Min-Suk Kong ; Young-Gy Shin |
| DOI |
http://dx.doi.org/10.6110/KJACR.2019.31.1.022 |
| Keywords |
신경망 ; 딥러닝 ; 전기 수요 ; 재실 정보 Neural network ; Deep learning ; Electricity consumption ; Occupancy information |
| Abstract |
Recently, numerous studies on the prediction of electricity consumption using deep-learning models have been conducted. The prediction models were mostly developed for a district scale since the influence of occupants' behaviors in such cases is small. On the other hands, the occupants generate huge uncertainty in predicting the future electricity demand. In this study, the unpredictable occupancy information was fed to a deep-learning model as a true value by assuming that in the future, the occupants may actively interact with the control systems using various smart device. The proposed model uses simple input values such as time of the day, base electricity load and occupancy information, while learning is achieved using measured data. Deep-leaning models with single and deeper layers were tested in this study. Both models showed excellent performance for data matching during the learning periods. The models also showed acceptable prediction performance for use in predictive control, with errors less than 30% in RMSE (cv). |