TU Delft Dream Team Epoch has achieved ninth place in their first competition of the year, the CityLearn Challenge. Engaged in a battle with 110 other teams from around the world, including Microsoft, the TU Delft students developed a new AI algorithm that contributes to better and smarter energy systems.

The CityLearn Challenge 2022 revolved around making the best use of green power from solar panels. By using an artificial intelligence (AI) model, energy consumption of consumers can be predicted. The model can then make the best choice at any time between storing or using the energy. At the same time, the algorithm takes into account energy prices and greenhouse gas emissions from power generation. In this way, AI can help consumers use energy with both the lowest price and also the fewest emissions.

The biggest challenge in the research phase came at the start of the competition, because it required the use of deep reinforcement learning. This is a technique for using machine learning in a complex environment, such as a power network, to make the optimal choices based on a predetermined reward mechanism. In this case, the choices are to charge or discharge the battery, and the reward mechanism is based on CO2 emissions and price. Since deep reinforcement learning has not been around very long, a lot of research had to be done on the latest techniques. In addition, it proved important to understand exactly how a network of houses and batteries was simulated in the CityLearn model.

The CityLearn Challenge is a prime example of the type of competitions Team Epoch students compete in. This AI team always chooses challenges that contribute to the UN Sustainable Development Goals. Their AI models thus contribute to sustainability, safety and health, for example. That’s how Epoch makes artificial intelligence more accessible to everyone and stimulates its uses for a truly better world.