Paper No 339 – Energy saving pre-cooling pattern search of an inverter air conditioner using a deep reinforcement learning algorithm – 13th IEA Heat Pump Conference, Jeju, Korea

In this study, we experiment with the energy saving pre-cooling operation of an inverter air conditioner using a deep reinforcement learning algorithm under a fixed going out time duration. An air conditioner with a rated cooling capacity of 6500 W is used. An indoor building load of 4500 W sensible heat and an outdoor temperature of 35?C are continuously applied throughout the study. During the going out time, three remote control temperature options (21, 26, 31?C) are given for simplicity, and a remote controlling pattern (set temperature sequences) that minimizes power consumption upon arrival home while satisfying the arrival target temperature condition is searched using deep reinforcement learning. For a given going out time duration, the power consumptions of the temperature operation pattern found by the computer deep reinforcement learning algorithm and a human are compared. Although the artificial intelligence searches for the power consumption operation pattern over a long period (7 ~ 9 days), depending on the learning environment, it identifies a more accurate and more energy saving and an unexpected operation pattern than by human analysis.

Download

Publication type Conf Proceedings Paper

Publication date 31 August 2021

Authors Myung Sup YOON, Won Sik YOON

Keywords Air conditioner; Deep reinforcement learning; Pre-cooling; AI; Energy saving; Smart HVAC; Set temperature sequences

Order nr Paper No 339

Download