Smart heat: Self-learning heating control system
Self-learning heating controls are on the rise. Researchers at Empa are convinced that building heating systems can "learn". In an experiment, they fed a new self-learning control system with weather data. As a result, the "smart" control was able to assess the behavior of the building and act predictively.
Heating control systems in factory buildings, airport terminals or high-rise office buildings are often already equipped with automated "predictive" technologies. These work with pre-programmed scenarios calculated specifically for the building and save operators a lot of heating energy. But for individual apartments and private houses, such individual programming is too expensive.
Last summer, a group of Empa researchers succeeded in proving for the first time that it can also be simpler: the intelligent heating and cooling control system does not necessarily have to be programmed; it can just as well learn to save by itself from the data of past weeks and months. Programming experts are no longer necessary. With this trick, energy-saving technology will soon be available for families and singles.
The decisive experiment took place in Empa's NEST research building. The "UMAR" (Urban Mining and Recycling) research unit offers optimal conditions for this: A large eat-in kitchen is symmetrically framed by two student rooms. Both rooms are 18 square meters in size. The entire window front faces east-southeast - towards the morning sun. In the UMAR unit, water runs through a stainless steel ceiling panel and provides the desired room temperature. The heating and cooling capacity can be calculated for the individual rooms via the respective valve position.
Cool smarter - thanks to weather forecast
Since project manager Felix Bünning and his colleague Benjamin Huber did not want to wait for the heating period, they started a cooling experiment as early as June 2019. The week from June 20 to 26 began with two sunny but still relatively cool days, then came a cloudy day, and finally the sun blazed over Dübendorf, chasing the outside temperature to just under the 40-degree mark. In the two bedrooms, the temperature should not exceed 25 degrees during the day and 23 degrees at night. A conventional thermostatic valve provided cooling in one room. In the other room, the experimental control system that Bünning and Huber had designed with their team operated. The artificial intelligence had been fed with data from the last ten months - and it knew the current weather forecast from MeteoSwiss.
More comfort with ¼ less energy
The result was extremely clear: The intelligent heating and cooling control system adhered much more closely to the com- fort specifications and required around 25 percent less energy to do so. This was mainly due to the fact that cooling was carried out in advance in the morning, when the sun was shining into the windows. The mechanical thermostat in the room opposite, on the other hand, only reacted when the temperature went through the roof. Too late, too frantically, and at full power. In November 2019, in a cool month with little sun, lots of rain and wind, Bünning and Huber repeated the experiment. Now the focus was on heating energy in the two rooms. At the time this issue went to press, the evaluation was still underway. But Bünning is sure that his predictive heating control will score points here, too.
He and his team have already prepared the next step: "To test the system in a real-world environment, we have planned a larger field trial in an apartment building with 60 apartments. We will equip four of these apartments with our intelligent heating and cooling control system." Bünning is already eager to see the results. "I believe that new machine learning-based controllers are a huge opportunity. Using this method, we can construct a good, energy-saving retrofit solution for existing heating systems with relatively simple tools and the data we collect."
