Experimental application of classification learning to generate simplified model predictive controls for a shared office heating system
The application of model-based predictive control (MPC) to commercial buildings has been a topic of increasing interest in the literature. Despite significant energy saving potential, commercial adoption has been limited. The literature suggests that the limited proliferation of MPC in buildings is due to a mismatch between the mathematical complexity of MPC and the expertise of practitioners. Accordingly, simplified MPC (sMPC) approaches may improve uptake. This paper applies an approach to MPC where rules are extracted from a detailed MPC algorithm to produce an algebraic model of fewer than 50 lines of code. The goal is to produce an sMPC that will be more accepted by practitioners. Testing in a shared office was executed over a 5-week period of the heating season from March 11 to April 14. The detailed MPC algorithm presented in this paper reduced heating energy consumption by 23% compared with conventional reactive control when tested under similar weather conditions. The sMPC algorithm achieved 95% of the heating energy savings of the detailed MPC algorithm with significantly less complexity. Based on this favorable result for a single case, further work should be conducted with larger sample sizes to conclusively determine if rule extraction can produce an effective sMPC.
|Journal||Science and Technology for the Built Environment|
Bursill, J. (Jayson), O'Brien, W, & Beausoleil-Morrison, I. (2019). Experimental application of classification learning to generate simplified model predictive controls for a shared office heating system. Science and Technology for the Built Environment. doi:10.1080/23744731.2018.1556052