Property
A private house with a floor area of 150 m² in Kyiv Region. The heating system is based on a Mitsubishi Electric Zubadan heat pump and water-based underfloor heating.
The property was selected as an experimental site for developing and testing artificial intelligence technologies in building services systems.
The brief
Modern heat pumps operate according to algorithms defined by the manufacturer. Even with weather compensation, they do not analyse the characteristics of an individual house, its thermal inertia, solar gains, equipment operating history or the outcomes of previous decisions.
The project aimed to create a system capable of analysing heat pump operation like an engineer and independently deciding on optimal heating operating modes.
Engineering approach
The engineering team behind SM Systems developed an experimental HVAC Intelligence platform that operates as a supervisory layer above the heat pump’s standard controls. The system does not replace the factory controller.
It analyses the controller’s operation, generates recommendations and manages temperature setpoints using its own decision-making logic.
Underfloor heating
Floor, indoor air and heating flow temperatures are analysed as a single thermal system.
Weather
The algorithm takes account of outdoor temperature, solar radiation and the forecast for the next 24 hours.
Energy
The heat pump’s current energy consumption is compared with its operating modes and the building’s response.
History
Every decision, setpoint change and building response is recorded for further analysis.
AI as a heating engineer
A key feature of the project was the integration of predictive models into operational analysis. At the end of each day, AI analyses the complete system operating log, just as an experienced heating engineer would.
It assesses temperature stability, underfloor heating behaviour, the building’s response to weather changes, the effectiveness of the selected temperature setpoints, the causes of unnecessary starts or extended operating cycles, and opportunities for further parameter optimisation.
Based on this analysis, AI produces recommendations for adjusting temperature setpoints and other thermal parameters of the system. The algorithm therefore not only controls the building but also gradually improves its own logic using accumulated data.
Prediction rather than reaction
Unlike most domestic control systems, the algorithm acts proactively. It takes account of forecasts for outdoor temperature and solar radiation, adjusting underfloor heating temperature setpoints in advance.
Before solar gains are expected to warm the building, the system can lower the heating water temperature in advance. Before a sharp drop in outdoor temperature, it can instead prepare the building for an increased heating load before the indoor temperature begins to fall.
This approach uses the building’s thermal inertia as an additional resource for maintaining comfort.
Continuous learning
During operation, the system continuously accumulates historical data. Every algorithm decision, temperature setpoint change, building response and heat pump energy consumption reading is recorded for further analysis.
This enables gradual improvement of the control algorithms based on the actual operation of a specific building, rather than laboratory models.
Result
The result was an experimental HVAC Intelligence platform for heating systems. It continuously analyses heat pump operation, predicts changes in the building’s heating load, takes account of weather forecasts and solar radiation, automatically controls temperature setpoints and records operating history.
The project demonstrated the feasibility of using AI as an engineering tool for developing modern HVAC systems.
What this project demonstrates
Most modern automation systems follow fixed algorithms that remain virtually unchanged after commissioning. We are exploring a different approach.
Control algorithms can evolve with the building, analyse the results of their own operation and continuously improve using accumulated data. We see this approach as the next stage in the development of building services systems.
Conclusion
This project became the SM Systems team’s research platform for HVAC Intelligence. It showed that artificial intelligence can be used not only for supervisory control or data visualisation, but also as an engineering tool capable of analysing HVAC operation, predicting building behaviour and supporting optimal decision-making.
We are now applying these technologies to commercial buildings, creating a new generation of climate control systems.
HVAC Intelligence transforms a heating system from a set of setpoints into an engineering process that learns from the building’s actual behaviour.