Photo: Empa

Hamstering cottages: AI for energy management in the house

The energy management in a house with a solar system is becoming more and more complex: When do I turn on the heating so that it is pleasantly warm in the evening? How much electricity can the hot water reservoir absorb? Is the energy still sufficient for the electric car? Artificial intelligence can help: Empa researchers have developed an AI control system that can learn all these tasks independently – saving more than 25 percent energy.

How were the old days simple: In the spring, when the heating oil prices fell, the tanks in the basement were simply filled to the brim. Then, until next season, all the worries were going on. Also for the car there was fuel at every corner. Around the clock. Full refuelling, ready to continue.

The exit from the fossil fuel economy makes it much harder for thrift foxes. Now energy prices no longer change annually, but hourly. Solar power is abundant at lunchtime – in the evening, the low sun hardly provides any energy, while returning commuters are rapidly increasing the demand for electricity in the city and the countryside. The effect can be seen so clearly on consumption graphs that scientists have given it its own name: “Duck-Curve”. When the duck raises its head, it becomes expensive for all those who now have to get electricity.

So looking at the clock when it comes to energy would be important for electric car drivers and homeowners. If you want to use the available renewable energy in a cheap and environmentally friendly way, you will no longer be able to rely on permanently installed thermostats and manually operated buttons.

A complex problem

Bratislav Svetozarevic, who conducts research in the Urban Energy Systems laboratory at Empa, has recognized the problem. What is needed is an automatic control system that hamsters energy at favourable times of the day and makes it usable for expensive times of the day. For example, the drive battery of your own car, which hangs at the charging station in the garage, could serve as storage. But Svetozarevic has to do with a complex problem: every house is different, and its inhabitants are. Depending on the weather and the season, the power generation of the solar systems as well as the need for heating or cooling capacity changes. An optimal energy control system must therefore learn the daily rhythm of a house and its inhabitants – and should also be able to react flexibly during operation, for example when a change in weather overturns all calculations.

Step one: the theory

The solution to such problems is artificial intelligence. The Empa researcher designed an AI control based on the Reinforcement Learning principle. If the system acts “correctly”, it receives a “reward”. Gradually, the controller perfects its behavior in this way.

Initially, the control was simulated only on the computer. The specifications: A certain room in a building had to be electrically heated to the desired temperature and hold it. At the same time, the system had to supply an electric car with electricity, which should be charged at least 60 percent at 7 a.m. in the morning and go on the journey. In the evening at 5 p.m., the electric car returns to the charging station with a residual charge and can also deliver electricity back to the house during the night hours. The control system was fed with weather data and room temperatures from last year and had to cope with two electricity tariffs: expensive electricity during the day between 8 a.m. and 8 p.m., cheap electricity during the night hours.

The result was astounding: the self-learning control saved around 16 percent of energy compared to a tightly programmed solution and also kept the desired room temperature much more precisely in the theory experiment.

Step two: Test in the real building

Now the controller had to pass the test in reality. Svetozarevic used NEST on the Empa campus. In the DFAB House unit, the AI algorithm controlled the temperature of a room for a week. At the same time, the 100 kWh storage battery in the NEST was used to simulate the battery of the electric car. This time, the result was even clearer: In a cool week in February 2020, the AI control system saved 27 percent of heating energy, compared to the neighboring student room, whose heating was operated with a permanently programmed (rule-based) control system.

“The beauty of our self-learning AI control system is that it can be used not only in the NEST research building, but also in any other building,” says Bratislav Svetozarevic. “You don’t need an engineer to program the control, or anyone who analyzes the house beforehand and calculates a tailor-made solution.”

Pleasant warmth in a economical way

In a next step, Svetozarevic and his colleagues now want to determine how the system can be expanded from a room to larger buildings. “In our first experiment, we wanted to depict a typical household of the future,” says the Empa researcher. For the sake of simplicity, the team has limited itself to heating and charging. The work, however, lays the foundation for much more. Svetozarevic is certain: “Our AI control system can still cope even when a photovoltaic system supplies electricity, a heat pump and a local hot water storage system have to be operated – and the comfort requirements of the residents change again and again.”

However, a new generation of electric cars is needed to use the AI system for optimal energy supply in the future. The current European and US models with the CCS quick-charging connection can only refuel, but not deliver one. Japanese cars with Chademo plugs, on the other hand, are designed for so-called bidirectional charging. Korean group Hyundai announced in December that it would also equip its new E-GMP electric car platform for bidirectional charging. This would allow electric cars to help save energy in the long term and at the same time stabilise the electricity grid.

More information on the subject can be found at: