Comfort control in residential housing using predictive controllers

Generalised Predictive Control (GPC) has shown to be an effective strategy for high performance applications, with good temporal and frequency properties (small overshoot, cancellation of disturbances, good stability and robustness margins). This paper presents an application of GPC to climate control in buildings, with particular attention focused on achieving strict temperature setpoint specifications. Specifically, the future temperature setpoint (step like in nature) is prespecified or known and the objective is to attain leading edge temperatures (positive increase) to within +/- 0.5 degrees C at the specified time. In addition, the controller must regulate the temperature to within +/- 0.5 degrees C through out the entire duration of the setpoint pulse and turn off the heater coincident with the trailing edge of the step change. In this paper, a new approach using GPC is developed for use in the comfort control problem. The final controller consists of a standard RST regulator derived from GPC and the development of an empirical anticipative rule which shifts the future setpoint backward in time in order to launch the heater sufficiently in the past to attain the leading edge setpoint value. An important result is that this new regulator is in operation continuously which is initialized only at start up. That is, no switching in or out of controllers is needed as is the case when PI with bang-bang is used. The validity and workability of such a predictive structure for thermal control in buildings is finally pointed out with the results of the predictive controller in the CLIM2000 environment, coupled with an automatic design of the tuning parameters for a simplified implementation.