LSTM-based Prediction of Building Power Consumption with PCA Based Data Dimension Change - Centered on a Campus Building in Daejeon -

PCA 적용 데이터 차원 변화에 따른 LSTM 기반 건물 전력 소비량 예측 - 대전 캠퍼스 건물을 중심으로 -

Building energy demand currently accounts for 30% of the total energy consumption, which has a great influence on the planning and operation of the energy market managed by energy suppliers. Furthermore, its importance has increased significantly with the advent of smart grid. Variables affecting building energy consumption include identified various environmental conditions that cast sophisticated effect on the energy performance of the buildings. However, due to a large number of potentially associated environmental variables, it is needed to extract embedded features so as to improve building energy prediction capability through adopting Principle Component Analysis which could reduce input data dimension. The primary objective of this study is to propose a high-precision building energy demand prediction model by reducing the dimensionality through PCA. Machine learning is implemented by using LSTM model, and prediction accuracy and performance are verified through R², RMSE, MAE, as well as computation time. The improvement ratio showed 14.93% increase when dimension-reduced dataset and normalized raw data were combined in comparison with the predicted case tested by using only normalized raw data. This study could support optimum building energy operation planning and design by promoting the creation and implementation of energy-efficient smart grid systems in the future.