Control temperature of room with reinforcement learning

Abstract

The human connection to the increase of average temperature on earth is a known issue because of the energy need that is partly full-filled with fossil fuels. Currently 40% of the world’s energy comes from buildings and by making heating/cooling systems more efficient there could be a big reduction of the energy need. The purpose of this research is to explore the possibilities of implementing machine learning to regulate temperature in a room. Using Python, Tensorflow and the Stable-baselines framework a simple model for reinforcement learning was created and trained on to explore if it was possible to use reinforcement learning to regulate the temperature in a room. The trained model managed to control the inside-temperature in a stable manner, this with a highly fluctuating outside-temperature and with a room-size never seen before. The paper will also discuss the steps taken to create a model, a working algorithm and further work.