논문집

원문다운로드
논문명 데이터 전처리 과정이 다층신경망을 이용한 건물 부하 예측 모델의 성능에 미치는 영향 분석/An Analysis of the Effect of the Data Preprocess on the Performance of Building Load Prediction Model Using Multilayer Neural Networks
저자명 성남철(Seong, Nam-Chul) ; 홍구표(Hong, Goo-Pyo)
발행사 한국건축친환경설비학회
수록사항 한국건축친환경설비학회 논문집  , Vol.16 No.4
페이지 시작페이지(273) 총페이지(12)
ISSN 1976-6483
주제분류 환경및설비
주제어 다층신경망; 데이터 전처리; 건물 부하; 예측 모델 Multilayer neural network; Data Pre-process; Building Load; Prediction Model
요약2 Data preprocess affects the performance improvement of predictive models using neural networks. However, previous studies have not analyzed the effects of data preproces. Therefore, in this study, the effect of data preprocess in a load prediction model using a multilayer neural network was analyzed. After constructing a multilayer neural network and generating learning data using a reference building, the performance of the predictive model was evaluated and compared through index of ASHRAE Guideline 14. As a result of the study, it was confirmed that the prediction results were improved through data preprocess. Compared to non-data preprocess models, the CvRMSE of models that performed data preprocess was improved from 21.47% to 10.74%, and MBE from 7.71% to 4.11%. It was found that the predictive performance improvement effect of data cleaning that removes missing values of data was greater than that of data transformation that converts the form of data.
소장처 한국건축친환경설비학회