27/05/2025 Hybrid AI for Building Management

Heating systems are among the largest energy consumers in buildings, and even small efficiency improvements can have a significant environmental and economic impact. By using the power of AI and historical data, the energy use can be made smarter and more sustainable, without sacrificing end users’ comfort.

Traditional thermostat systems often follow rigid schedules, unable to adapt to changing factors like weather or occupancy patterns. This inflexibility often leads to unnecessary energy consumption or discomfort for occupants. But what if your thermostat could learn from the past to make more efficient decisions?
Energy Efficient Control with AI

Using a type of AI called model-free Deep Reinforcement Learning (DRL), Sara Ghane from IDLab-Antwerp developed a system that optimizes thermostat control for heating systems, balancing energy savings and comfort. The DRL method learns a decision-making control strategy using historical data rather than real-time interactions. This approach ensures uninterrupted comfort during the training phase.
How It Works

The system is built on a type of offline DRL. It learns to adjust thermostat setpoints based on historical data from HomeLab, including indoor and outdoor temperatures, day/night status, and heat pump energy use. By designing the model to prioritize both energy use and thermal comfort, this approach dynamically adapts to changing conditions.
Results from HomeLab

The tests in HomeLab showed impressive results:

  • Around 18% energy savings compared to the rule-based method.
  • Reliable indoor temperature management, meeting comfort standards even during cold weather.
This reduction in energy use translates to significant cost savings and ultimately lower GHG emissions. Therefore, these results demonstrate the potential of DRL-based systems to make building energy management more efficient by learning from past data—without requiring a pre-built model of the environment.
The Future of Energy Efficient Thermostats

Looking ahead, this system has the potential to scale across multiple buildings by incorporating diverse datasets from different buildings. Innovations like this, lay the groundwork for smarter, energy-efficient technologies that align with UN sustainability goals and enhance everyday comfort for building occupants