Paper Conference

Proceedings of ASim Conference 2016: 3rd Asia conference of IBPSA-China, Japan, Korea

     

A REVIEW ON DATA-DRIVEN MODELS FOR MUTI-TIMESCALE BUILDING ENERGY PREDICTION

Y. Pan, M. Zhu, Z. Huang

Abstract: Data-driven Models (DMs) play a significant role in the building energy prediction researches. The lack of targeted summary on DMs usually makes researchers and scientists engaged in promoting building energy performance confused of the selection and application of DMs. This paper primarily summarizes the application of DMs in the field of building energy prediction. According to the methodology of DMs development, the Regression Model (RM), Time Series Model (TSM), Genetic algorithm (GA), Artificial Neural Network (ANN)/ Support Vector Machine (SVM) and Calibrated Simulation (CS) are classified into DMs. From the view point of the model application, the function of the prediction load/energy achieved by a specific DM depends on its time scale. Under different building management goals for different levels of decision-makers and researchers, the model selection, the data requirements and applicable objects are all distinct accordingly. Different from previous reviews that mainly pay attention to the methodology and the performance comparison of building energy prediction models, we review the DMs' model development and application in the multi-timescale building energy prediction, including hourly/daily, monthly, annual and mutidecadal timescale. This paper will contribute to reasonable selection of existing DMs on the basis of the building operation stage, the available monitored data quality, the building energy prediction requirement, and the building energy efficiency goal for an object building or a building type.
Keywords: Data-driven Models (DMs), building energy prediction, multiple timescales, model application
Paper:
asim2016_333