Rank: 1st Place
Winning 1st Place in Energy Consumption Forecasting at Datathon Zurich
About the Competition
Team Epoch recently achieved a major milestone by winning 1st place at the ETH Zurich Datathon 2026. Competing against 60 elite teams, including academic researchers and industry professionals, the team outperformed the baseline by nearly 50%. This victory was secured during a 24-hour continuous coding phase starting on Sat, Apr 18, 2026. The winning team members included Rein Viegers, Willem Dieleman and Maxim Cardenas from Team Epoch and Rafael Alani a masters student at TU delft.
Relevance
The challenge required participants to forecast short-term energy consumption for retail clients in Spain. This is a critical problem for energy suppliers, as accurate demand predictions are essential to manage purchasing upstream, mitigate imbalance risks, and protect profit margins. When actual consumption deviates from predicted demand, the resulting forecast errors translate directly into significant financial costs for the energy supplier. Therefore, achieving highly accurate prediction, as demonstrated by the winning solution, is crucial for operational stability and maintaining profitability in the volatile energy sector.
Analytics Club at ETH Zurich
The Analytics Club at ETH Zurich is the host of the ETH Zurich Datathon 2026. This event is a highly competitive, top-tier Swiss datathon that provides a rigorous environment where elite teams, including academic researchers and industry professionals from leading institutions, compete. The competition focuses strictly on advanced statistical modeling, predictive accuracy, and analytical explainability.
The challenge was sponsored by Axpo Group and Databricks. Axpo Group, a major energy supplier, provided the core problem: forecasting short-term electricity demand for retail clients in Spain.The technical solution was built and deployed using the Databricks platform, which supports the entire machine learning workflow—from building the sophisticated ensemble of models to facilitating online training, integrating new data into a “Lakehouse,” and deploying solutions to Databricks Apps using tools like MLflow.
Technical Details
The team’s technical approach involved building a sophisticated ensemble of models using the Databricks platform, including Chronos2, LSTMs, and GBDTs. A key part of their strategy was modeling all 18 regions of Spain separately to capture localized consumption patterns. They also integrated external data such as weather forecasts and a custom-built Iberian holiday calendar to improve predictive accuracy.
In terms of results, the team delivered an average reduction of ~12.7 MW in forecast errors, which translates to estimated savings of over $1.67M in imbalance market costs over a three-month period. Beyond their technical performance, the jury was impressed by the team’s focus on scalability and real-world deployment during their final pitch.
UN Sustainable Development Goals
This project supports UN Sustainable Development Goal 7: Affordable and Clean Energy. By providing accurate energy consumption forecasts, the solution helps energy suppliers optimize resource management, reduce waste, and improve the efficiency and reliability of energy distribution networks.
