TELUS AI Agent Achieves Energy Savings via Reinforcement Learning
TELUS successfully tested an AI agent using reinforcement learning to optimize energy consumption in a data center as part of its Energy Optimization System (EOS) project. Developed in collaboration with the Vector Institute for Artificial Intelligence, the agent was tasked with managing cooling systems to meet sustainability goals, specifically aiming to reduce energy intensity by 50% between 2020 and 2030. Unlike traditional supervised learning, this reinforcement learning model learned through interaction with the environment, considering both immediate and long-term effects of its actions. In simulator tests, the agent achieved a 2%-15% reduction in energy use by innovatively allowing temperatures to rise closer to thermal limits before stabilizing, rather than maintaining excessive buffers. It strategically utilized free cooling from outside air versus energy-intensive compressors based on weather forecasts. This pilot marks a significant step in applying AI for industrial sustainability, demonstrating how goal-oriented algorithms can autonomously discover efficient operational strategies for complex physical systems like HVAC units in real-world settings.
Wire timeline
TELUS AI Agent Achieves Energy Savings via Reinforcement Learning
TELUS successfully tested an AI agent using reinforcement learning to optimize energy consumption in a data center as part of its Energy Optimization System (EOS) project. Developed in collaboration with the Vector Institute for Artificial Intelligence, the agent was tasked with managing cooling systems to meet sustainability goals, specifically aiming to reduce energy intensity by 50% between 2020 and 2030. Unlike traditional supervised learning, this reinforcement learning model learned through interaction with the environment, considering both immediate and long-term effects of its actions. In simulator tests, the agent achieved a 2%-15% reduction in energy use by innovatively allowing temperatures to rise closer to thermal limits before stabilizing, rather than maintaining excessive buffers. It strategically utilized free cooling from outside air versus energy-intensive compressors based on weather forecasts. This pilot marks a significant step in applying AI for industrial sustainability, demonstrating how goal-oriented algorithms can autonomously discover efficient operational strategies for complex physical systems like HVAC units in real-world settings.
Vector Institute for Artificial Intelligence