EPFL, Empa, and SBB Develop AI Model for Improved Electricity Demand Forecasting for Swiss Rail
A new AI model developed by researchers from EPFL and Empa in collaboration with Swiss Federal Railways (SBB) significantly reduces electricity demand forecasting errors. The model combines historical data with future operational and timetable information.
Researchers from EPFL and Empa, working with Swiss Federal Railways (SBB), have developed an artificial intelligence model that substantially improves next-day electricity demand forecasts for the entire SBB rail network.
The model's key innovation lies in combining historical energy consumption data with contextual information, including train timetables, operational planning data, and weather forecasts. The published results indicate an average prediction error reduction of 26.6%, with major errors on unusual operating days reduced by approximately 80%.
Accurate electricity demand forecasting is critical for Switzerland's rail system, which operates on a dedicated power grid. Forecasting errors can lead to operational risks and increased costs. Research suggests that even a 1% reduction in forecasting error could yield annual savings of around one million Swiss francs.
The research team proposes that similar methods could be applied to other sectors, such as building energy systems, manufacturing, and logistics and supply chain management. By aggregating fragmented information from various sources and planning for future operations, demand can be predicted more precisely.