Abstract:

We propose a method for model-based control of building air conditioning systems that minimizes energy costs while maintaining occupant comfort. The method uses a building thermal model in the form of a thermal circuit identified from collected sensor data, and reduces the building thermal dynamics to a Markov decision process (MDP) whose decision variables are the sequence of temperature set-points over a suitable horizon, for example one day. The main advantage of the resulting MDP model is that it is completely discrete, which allows for a very fast computation of the optimal sequence of temperature set-points. Experiments on thermal models demonstrate savings that can exceed 50% with respect to usual control strategies in buildings such as night setup.