Paper No 236 – Application of a deep reinforcement learning algorithm in household inverter air-conditioner temperature control – 13th IEA Heat Pump Conference, Jeju, Korea
A deep reinforcement machine learning algorithm was applied to household inverter air-conditioner precision temperature control. Generally, air-conditioner temperature control aspects rely on the specific technology of the product, cooling room area size, outdoor temperature variation, and indoor building load variation when the set temperature is fixed. In this study, we fixed the test product and room size, and used the given variations of outdoor temperature and indoor building load over the course of one day. Even though the test product showed satisfactory performance at the remote controller set temperature of 26?C without a machine learning algorithm, we experimented with deep reinforcement learning performances to check whether the test product could follow general product performances or surpass the product ability in the precision temperature control by applying only high and low set temperatures alternatively. Training started with no disturbances of constant building cooling load and outdoor temperature. It resulted in reasonably accurate temperature control with the real training environments of Korean and Middle Eastern climates