Supply temperature control of a heating network with reinforcement learning

Heating networks are typically controlled by a heating curve, which depends on the outdoor temperature. Currently, innovative heating networks connected to low heat demand dwellings ask for advanced control strategies. Therefore, the potentials of reinforcement learning are researched in a heating network connected to a central heat pump and four dwellings. The comparison between a discrete and continuous action space is made with respect to the weight factor of the reward function. The results indicate that in both cases the reinforcement learning-based controlling of the supply temperature can generally ensure energy savings while keeping the occupant's temperature requirements in comparison to the rule-based controller.

Automated summary

Heating networks are typically controlled by a heating curve that is dependent on outdoor temperature, but innovative networks connected to low heat demand dwellings require advanced control strategies. Research has shown that reinforcement learning can lead to energy savings while meeting occupants' temperature requirements.