Telus and the Vector Institute for Artificial Intelligence last week announced the launch of the Energy Optimization System (EOS), with the aim to help reduce operational costs and minimize electricity use in commercial buildings, such as datacentres, across Canada.
This collaborative development uses model-based reinforcement learning (MBRL) to fine tune the heating, ventilation, and air conditioning (HVAC) systems across network locations, allowing for energy-efficient temperature control.
“In MBRL, we train an agent to learn from the environment it resides in, and based on a reward/penalty system, the agent selects the most appropriate action to execute,” Jaime Tatis, vice-president, Data Strategy and Enablement at Telus, told Channel Daily News.
Annually, an estimated 40 per cent of the energy consumed across Telus network locations is directed towards cooling telecommunications equipment, which is critical to maintaining optimal network performance. Telus team members collaborated with the Vector Institute to build and test this solution to reduce the number and create a more sustainable use of HVAC equipment in 24/7 datacentres.
“For our system, we used a complete year of data to train and simulate the environment, which allowed us to develop a good understanding of how it will react in the live environment,” said Tatis. “The advantage of this approach is that temperature is a slow-moving variable, but with a simulator, we can train it quickly and therefore confidently deploy it in a critical live environment more efficiently.”
By optimizing HVAC systems with this new artificial intelligence (AI) technology, it is possible to achieve significant energy savings in datacentres and other network locations, reducing the overall environmental impact. Results from a pilot test showed a decrease of almost 12 per cent in reduced annual electricity consumption in a small datacentre. These results are based on a summer test completed in a small Telus datacentre (equipment room) in Cambridge, ON from June to July 2021 and a winter test done in the same centre in Jan. 2022, said Tatis.
In addition, the algorithm considers the weather forecast to help it make a decision on when to run cooling (either more expensive compressor cooling or less expensive free cooling) or heating to ensure a consistent temperature and better regulate temperatures during shoulder seasons.
Telus and Vector say they have elected to open source this new algorithm as a contribution to the energy conservation community. More information on how the algorithm works can be found here.
“After three years of working on EOS with a focus on using AI for good, this partnership recognized the value of open sourcing the MBRL algorithm for cost reduction and to create efficiencies for other industries and organizations,” said Deval Pandya, Director of AI Engineering at Vector. “This is a brilliant example of how, together, our expertise in research and engineering can create value and make it easier to deploy leading AI research outcomes. Now we want to amplify the project’s value by open sourcing it for others to adopt.”
“At TELUS, we are making choices and taking actions on our commitment to protect our planet. We are working to achieve our goal to be net carbon neutral by 2030, while also procuring 100 per cent of our electricity requirements from renewable or low-emitting sources by 2025,” said Tatis.
