How AI is optimising HVAC

As organisations work to reduce their energy consumption and associated carbon emissions, indoor heating and cooling is an area that needs to become more sustainable.

Currently, HVAC — heating, ventilation and air conditioning — represents, on average, about 40% of a building’s total energy use, according to analysis by the International Energy Agency. Methods that conserve electricity while still providing a comfortable indoor environment for workers could make a significant difference in the fight against climate change.

Now, researchers from Osaka University have devised an AI-driven algorithm for controlling HVAC systems that demonstrates significant energy savings. This method does not require complex physics modelling, or even detailed previous knowledge about the building itself, the scientists said.

During cold weather, it is sometimes challenging for conventional sensor-based systems to determine when the heating should be shut off. This is due to thermal interference from lighting, equipment or even the heat produced by the workers themselves. It can lead to the HVAC being activated when it should not be, wasting energy.

To overcome these obstacles, the researchers employed a control algorithm that worked to predict the thermodynamic response of the building based on data collected. According to the researchers, this approach can be more effective than attempting to explicitly calculate the impact of a multitude of complex factors that might affect the temperature, such as insulation and heat generation. Here, the HVAC control system was designed to ‘learn’ the symbolic relationships between the variables, including power consumption, based on a large dataset.

“Our autonomous system showed significant energy savings, of 30% or more for office buildings, by leveraging the predictive power of machine learning to optimise the times the HVAC should operate,” said lead author Dafang Zhao. “Importantly, the rooms were comfortably warm despite it being winter.”

The algorithm worked to minimise the total energy consumed, the difference between the actual and desired room temperature, and the change in the rate of power output at peak demand. “Our system can be easily customised to prioritise energy conservation or temperature accuracy, depending on the needs of the situation,” added senior author Ittetsu Taniguchi.

The researchers noted that their approach may see rapid adoption during times of rising energy costs, making their findings good for both the environment and the bottom line.

The team’s article, ‘Data-driven Online Energy Management Framework for HVAC Systems: an Experimental Study’, was published in the journal Applied Energy.

Image credit: iStock.com/sturti