New underfloor heating solutions requires a new form of control strategy, as the actuation system is very different from actuators in traditional control systems present in traditional houses today. In addition to the new actuation system, it is widely recognized that room temperature control in underfloor heating systems often does not meet today's comfort requirements, especially in systems where the underfloor heating is embedded in concrete floors.
In this project Deep Reinforcement Learning together with information from weather forecasts are investigated, with the aim to be able to control the underfloor heating system and improve temperature control without prior knowledge of the building. This provides a Plug-and-Play underfloor heating system, which is capable of improving comfort and minimizing energy consumption in any house installation. Proximal Policy Optimization (PPO) is currently being deployed with a Multi-Agent Reinforcement Learning (MARL) architecture.
