Predictive HVAC Control Using a Markov Occupancy Model

This paper presents a model predictive control (MPC) technique for building heating, ventilation, and air conditioning (HVAC) systems. It incorporates the building's thermal dynamics, local weather predictions, and a stochastic occupancy model to reduce energy consumption while maintaining occupant comfort. Using approximate dynamic programming and a cost function weighted by expected occupancy, the scheme extends the capability of conventional model predictive control by pre-conditioning thermal zones before occupancy begins and reducing conditioning before occupancy ends. The resulting control law may be synthesized step-wise using an on-line optimization or may be periodically synthesized off-line and downloaded into an embedded controller. Simulation results demonstrate the efficacy of both approaches.