Advanced Optimal Control of Indoor Environmental Devices for Indoor Air Quality Using Reinforcement Learning

This study aims to develop and evaluate an advanced control method for acceptable indoor air quality (e.g., particulate matter and CO2) with low energy consumption in a residential space. A ventilation system, an air purifier, and a kitchen hood system are installed in the testbed to maintain a healthy IAQ. To accomplish the objective, we use a double deep Q-network (DDQN) which is one of the reinforcement learning. This study utilizes a co-simulation platform with EnergyPlus and Python. The optimal control model was trained for 5 days to represent various outdoor conditions and indoor living contexts in residential buildings by introducing emission rates of the indoor fine particles according to occupant’s activities. The evaluation of the suggested optimal control was performed by comparison with a simple on/off method for environmental devices. As a result, the DDQN control showed an improvement of 2.5% (PM 2.5) and 0.6% (CO2) of healthy air ratio while reducing 45.5% of energy consumption.