Energy optimization associated with thermal comfort and indoor air control via a deep reinforcement learning algorithm

The aim of this work is to propose an artificial intelligence algorithm that maintains thermal comfort and air quality within optimal levels while consuming the least amount of energy from air-conditioning units and ventilation fans. The proposed algorithm is first trained with 10 years of simulated...

Full description

Saved in:

Bibliographic Details
Published in:	Building and environment Vol. 155; pp. 105 - 117
Main Authors:	Valladares, William, Galindo, Marco, Gutiérrez, Jorge, Wu, Wu-Chieh, Liao, Kuo-Kai, Liao, Jen-Chung, Lu, Kuang-Chin, Wang, Chi-Chuan
Format:	Journal Article
Language:	English
Published:	Oxford Elsevier Ltd 15-05-2019 Elsevier BV
Subjects:	Air conditioners Air conditioning Air conditioning equipment Air quality Algorithms Artificial intelligence Carbon dioxide Computer simulation Control methods Deep reinforcement learning Energy consumption Indoor air pollution Indoor air quality Indoor environments Levels Machine learning Optimization Subtropical zones Thermal comfort Ventilation Ventilation fans Air conditioning Deep reinforcement learning Ventilation Thermal comfort Optimization Indoor air quality
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	The aim of this work is to propose an artificial intelligence algorithm that maintains thermal comfort and air quality within optimal levels while consuming the least amount of energy from air-conditioning units and ventilation fans. The proposed algorithm is first trained with 10 years of simulated past experiences in a subtropical environment in Taiwan. The simulations are carried out in a laboratory room having around 2–10 occupants and a classroom with up to 60 occupants. The proposed agent was first selected among different configurations of itself, with the 10th-year of training data set, then it was tested in real environments. Finally, a comparison between the current control methods and this new strategy is performed. It was found that the proposed AI agent can satisfactorily control and balance the needs of thermal comfort, indoor air quality (in terms of CO2 levels) and energy consumption caused by air-conditioning units and ventilation fans. For both environments, the AI agent can successfully manipulate the indoor environment within the accepted PMV values, ranging from about −0.1 to +0.07 during all the operating time. In regards to the indoor air quality, in terms of the CO2 levels, the results are also satisfactory. By utilizing the agent, the average CO2 levels fall below 800 ppm all the time. The results show that the proposed agent has a superior PMV and 10% lower CO2 levels than the current control system while consuming about 4–5% less energy. •An artificial intelligence algorithm (AI) is developed for thermal comfort, air quality, and energy consumption.•The developed algorithm is applicable for subtropical environment with cooling-only demand.•The simulations are in line with the experimental results in a laboratory room and a classroom.•The AI agent shows better thermal comfort and much lower CO2 levels than those without AI agent.•AI reveals 4–5% lower energy consumption with superior thermal comfort and 10% lower CO2 levels.
ISSN:	0360-1323 1873-684X
DOI:	10.1016/j.buildenv.2019.03.038