데이터 기반 모델을 위한 필터링 기법의 적용

Recently the data-driven model has been highlighted due to its prediction capability, reliability and accuracy. It differs from first-principle based simulation model in that it predicts the system's behavior based on the relationship among the measured data. With this in mind, the authors present in this paper RANdom SAmple Consensus (RANSAC), Wavelet Transform, and Gaussian Process techniques which can define correlated data pairs out of large time-series Building Energy Management System (BEMS) data. This paper addressed the validation of the aforementioned techniques in terms of improved performance of the data-driven model (Gaussian Process Model).