논문집

원문다운로드
논문명 입력변수의 불확실성 보정을 위한 인공신경망 대리모델 기반의 민감도 분석 및 베이지안 MCMC/Sensitivity Analysis and Bayesian MCMC based on Artificial Neural Network Surrogate Model for Input Variable Uncertainty Calibration
저자명 박성철(Park, Seong-Cheol) ; 하주완(Ha, Ju-Wan) ; 박경순(Park, Kyung-Soon) ; 송영학(Song, Young-Hak)
발행사 한국건축친환경설비학회
수록사항 한국건축친환경설비학회 논문집  , Vol.15 No.4
페이지 시작페이지(326) 총페이지(12)
ISSN 1976-6483
주제분류 환경및설비
주제어 입력변수; 불확실성; 민감도 분석; 대리모델; 베이지안 보정; MCMC Input Variable; uncertainty; Sensitivity analysis; Surrogate model; Bayesian calibration; Monte-carlo Markov-chain
요약2 To have a precise simulation of building and quantitative performance evaluation, it is necessary to have high accuracy of input variables to implement a simulation. Inherent uncertainties (unknown fault or rapid error generation, etc.) in input variables can have a significant impact on energy performance evaluation. To solve this, this study conducted Artificial Neural Network (ANN) surrogate model-based Sobol sensitivity analysis using data of water-cooled variable refrigerant flow air conditioning system and Bayesian Markov chain Monte Carlo (MCMC) using no-U-turn sampler (NUTS) algorithm was performed. The main results showed that the Coefficient of the Variation of the Root Mean Square Error of the ANN surrogate model was 23.4%, and the top 6 variables that affected the energy usage through the sensitivity analysis results were selected. Then, the prior and posterior probability distributions were plotted through Bayesian MCMC. The convergence verification result of MCMC verified that the mean R-hat value was 1.0, which ensured the accuracy of MCMC. For future study, a virtual sensor will be developed to detect and diagnose faults in building heating, ventilation, and air conditioning based on the study results.
소장처 한국건축친환경설비학회