Data requirements and performance evaluation of model predictive control in buildings: A modeling perspective

Abstract

Model predictive control (MPC) has shown great potential in improving building performance and saving energy. However, after over 20 years of research, it is yet to be adopted by the industry. The difficulty of obtaining a sufficient control-oriented model is one major factor that hinders the application. In particular, what data is required to build the model and what control performance can be expected with a certain model remain unclear. This study attempts to uncover the underlying reasons and guide future research to tackle the challenges. It starts by clarifying a finer categorization of past studies with respect to both modeling methods and modeling purposes. An extended Level of Detail (LoD) framework is proposed to quantify the data usage in each study. Accordingly, meta-analyses are conducted to compare the data requirements of different modeling categories. The criteria and approaches for model performance evaluation are summarized and classified into validation and verification methods, followed by a discussion about the relationship between the model and control performance. The critical review provides new perspectives on the data requirements and performance evaluation of control-oriented models. Ultimately, the paper concludes with five directions for future research to bridge the gaps between data requirements, model performance, and control performance.