Reinforcement Learning based optimisation of heating curve

Flow temperature adjustment by means of Q-learning

Abstract

In this contribution the potential of an intelligent supply  temperature control for a heating network of a modern  office building is analysed. As the building is equipped with  a low temperature floor heating system and large window  areas, overheating may occur during days with significant  amount of solar energy but cool nights. With the  optimisation of the supply temperature using a reinforcement learning based approach – Q-learning – overheating can effectively be reduced compared to a  standard heating curve, leading to an improvement of  comfort. A recommendation system is used to integrate the  new controller into the real building.