A group of researchers from the Republic of Korea are exploring intelligent control methods such as quantum reinforcement learning (QRL) based on quantum computing principles for HVAC systems.
This will significantly accelerate the machine learning process while handling the complexity of real-world building dynamics.
Residential heating, ventilation, and air conditioning (HVAC) systems constitute a significant proportion of energy usage in buildings, necessitating energy management optimisation. Occupancy aware HVAC control is a promising option with 20-50% energy savings in homes. However, occupancy sensing technology suffers from long payback times, privacy issues, and poor comfort.
Moreover, there is an increasing need for advanced technologies that help regulate indoor air quality alongside energy control.
Dr. Sangkeum Lee, Assistant Professor of Computer Engineering at Hanbat National University, Korea, has presented the first demonstration of continuous-variable, quantum-enhanced reinforcement learning for residential HVAC and home power management.
The findings were published in Volume 21 of “Energy and AI” last month.
"Unlike conventional reinforcement learning techniques, QRL leverages quantum computing principles to efficiently handle high dimensional state and action spaces, enabling precise HVAC control in multi-zone residential buildings,” Lee said.
“Our framework integrates real-time occupancy detection using deep learning with operational data, including power consumption patterns, air conditioner control data, and external temperature variations.”
Furthermore, the proposed technology integrates features such as multi-zone cooling (control temperature of individual zones in building) and clustering (group similar data points and adjust cooling). Thus, a single controller jointly optimises comfort, energy cost, and carbon signals in real time.
The researchers performed simulations based on real world data from 26 residential households over a three-month period.
QRL HVAC control outperformed deep deterministic policy gradient method and proximal policy optimisation algorithm, maintaining thermal comfort, while achieving 63% and 62.4% reductions in power consumption, and 64.4% and 62.5% decrease in electricity costs, respectively.
This approach is retrofit-friendly and works with standard temperature, occupancy and carbon dioxide sensors, common HVAC equipment and thermostats.
It is also robust to uncertainty, handling noisy forecasts on weather and occupancy, and device constraints. In addition, the general framework can be extended from apartments to small buildings and microgrids.
"It can be utilised in smart thermostats and autonomous home energy management systems that co-optimize comfort, bills, and emissions without manual tuning and rooftop photovoltaics and home battery scheduling,” Lee said.
“Our framework is also useful for utility demand-response and time-of-use programs with automated control."
QRL based HVAC control can be applied at community or campus scale through grid-interactive efficient buildings and virtual power plants (VPPs). Herein, millions of homes can coordinate as VPPs to stabilise renewables-heavy grids.
It can also ensure personalised indoor environmental quality within carbon budgets and integrate advanced intelligent control options.
As hardware matures with time, quantum-accelerated policy search could accelerate training for complex multi-energy systems such as HVAC, electric vehicles, and energy storage systems.
Eventually, this work will pave the way toward standardised secure controllers that can be certified and deployed at a wide scale, he said.