FETS Benchmark Tests Foundation Models for Energy Forecasting

A new benchmark called FETS (Foundation Models in Energy Time Series Forecasting) evaluates whether foundation models can outperform dataset-specific machine learning in energy time series forecasting. The benchmark provides a structured overview of energy forecasting use cases across three dimensions: stakeholders, attributes, and data categories. It collects and analyzes 54 datasets across 9 data categories. The work addresses the gap in applying foundation models to energy forecasting, which remains largely unexplored despite their success in other prediction tasks. The transition to a climate-neutral energy system drives the need for accurate forecasting, but current dataset-specific approaches limit scalability and require extensive training data. The FETS benchmark aims to enable generalizable patterns through extensive pretraining.