Artificial intelligence could help cool buildings more efficiently. London-based research firm DeepMind trained an AI to minimise energy usage while controlling building cooling systems under different weather conditions.
鈥淎 key benefit of using AI in a [building] climate control system is that it is constantly monitoring and adapting to the changing world,鈥� says at DeepMind.
DeepMind worked with Google and Trane Technologies, a manufacturer of building control systems, to develop an AI for controlling cooling systems at university campus buildings and a mixed-used building with apartments, restaurants and shops.
Advertisement
The team trained the AI to continuously search for actions that would minimise overall energy usage without sacrificing the comfort of people inside. The AI was first trained using less than a year鈥檚 worth of historical data from the standard control system for each building. It then learned further by controlling the buildings鈥� chiller plants while monitoring the effects, as well as weather patterns and changing levels of demand for cooling. The AI used what it had learned to generate and assign scores to possible actions for every decision that it faced.
The system helped save between 9 and 13 per cent on energy required for cooling over three months spanning the summer and autumn of 2021.
Most buildings still rely on simple control systems that require humans to manually adjust the settings and cannot adapt automatically to changing factors such as weather conditions, says at the Pacific Northwest National Laboratory in Washington state. 鈥淏uilding control has been considered a very boring field, not really sexy like drones or computer games,鈥� he says.
Cooling homes and buildings accounts for about 10 per cent of the world鈥檚 energy demand. So even though the individual building energy savings were relatively modest, they could have a significant effect if rolled out on a mass scale.
However, that won鈥檛 be easy. A lot of human labour had to go into helping the AI understand the specific control systems in the different buildings, says Luo. That required lengthy conversations with the building managers to create learning shortcuts by pre-programming certain knowledge into the AI, he says.
It is also unclear how well DeepMind鈥檚 AI system would handle a building鈥檚 entire system beyond the water-based cooling component, says at BrainBox AI, a Canadian company that builds and deploys similar technology. BrainBox AI claims to have slashed building energy costs by up to 25 per cent through controlling the entire heating, ventilation and air conditioning system.
Owners of older buildings with limited budgets might achieve comparable or greater energy savings by retrofitting the buildings with better insulation or more energy-efficient heating and cooling systems instead of buying AI-powered building controls, says at the National Renewable Energy Laboratory in Colorado. But figuring that out would require more data and test demonstrations.
Reference:
