A practical artificial intelligence-based approach for predictive control in commercial and institutional buildings

Abstract

This paper presents a methodology for the development and implementation of Model Predictive Control (MPC) in institutional buildings. This methodology relies on Artificial Intelligence (AI) for model development. An appropriate control-oriented model is a critical component in MPC; model development is no easy task, and it often requires significant technical expertise, effort and time, along with a substantial amount of information. AI techniques enable rapid development and calibration of models using a limited amount of information (i.e. measurements of few variables) while achieving relatively high accuracy. In this study, the MPC algorithm targets the reduction of natural gas consumption by optimizing the transition between night set-back and daytime indoor air set-point values as a function of the expected weather. This MPC strategy was implemented in an institutional building in Varennes (QC), Canada, during the heating season 2018–19. A significantly better performance was achieved when compared with “business as usual” control strategies: the natural gas consumption and greenhouse gas (GHG) emissions were reduced by approximately 22%, and the building heating demand by 4.3%. The proposed strategy is scalable and can be replicated in other buildings.