Experimental investigation of model predictive control-based rules for a radiantly cooled office

Abstract

A novel hybrid approach for supervisory control was applied to a test cell with thermo-active building systems and fan-assisted natural ventilation. Model predictive control was first used to identify combined thermo-active building systems and ventilation control strategies that maximized cooling energy savings while preserving thermal comfort. A rule extraction process using classification and regression trees then yielded supervisory rules capable of reproducing nearly all of the energy and comfort benefits of the model predictive control solutions when simulated. An experimental test of the rules was conducted on the same facility, yielding 40% average cooling energy savings compared to a base case, with comparable comfort. A variety of model input mismatches, including weather, model parameters, internal gains, and imperfect weather forecasts, degraded the performance of the rule significantly under experimental conditions, and analysis suggests that energy savings could have been nearly double had these factors been eliminated.