HVAC (heating, ventilation, and air conditioning) accounts for a sizable portion of the world’s CO2 emissions.

Another fact: Space cooling itself accounts for around 10% of global electricity demand. HVAC efficiency gains can therefore make a dent in the efforts needed to control climate change and global warming. These gains can be boosted by employing AI technology (such as Reinforcement Learning) and data analysis that can make fine tuning adjustments to the HVAC equipment in real time. (MARKTECHPOST)

DeepMind, the British artificial intelligence subsidiary of Alphabet Inc. (NASDAQ: GOOGL) and research laboratory founded in 2010, has published a study on the use of reinforcement learning for controlling commercial cooling systems.

Reinforcement Learning

Reinforcement learning (RL) is a machine learning training method based on rewarding desired behaviors and/or punishing undesired ones. In general, a reinforcement learning agent is able to perceive and interpret its environment, take actions and learn through trial and error.

DeepMind used RL that relied on previous work regulating the cooling systems of Google’s data centers, to increase the energy efficiency of HVAC control in two commercial buildings.

It conducted live experiments on these two real-world facilities in partnership with Trane Technologies, a building management system provider.

The paper details the experience of using RL to provide real-time supervisory setpoint recommendations to the chiller plant (an essential and energy-intensive part of many HVAC systems) in the buildings.

The researchers write:

“This work has presented us with an array of challenges, from common ones such as expensive and noisy data to more novel ones such as having different operating modes and multi timescale dynamics. In order to mitigate these challenges, we employed a combination of general RL solutions and domain-specific heuristics.

“The resulting system was benchmarked against heuristics-based controllers provided by Trane, and showed a 9-13% reduction in energy use while satisfying system constraints.”

The key payoffs from the research:

• It shows that the same underlying RL algorithm may be used for controlling the cooling systems of several different large commercial buildings, and can improve upon industry-standard heuristics-based controllers in doing so.
• It identified concrete examples of RL challenges and describes how these challenges were addressed in a real system.
• Finally, it sheds light on what might be needed to further scale RL for HVAC control, providing a potential blueprint for unlocking more efficiency and scaling to new use cases in the future.

Related Story:  Global X Launches Green Building ETF  

Image by F. Muhammad from Pixabay