FETS Dataset: Foundation Models Enable Scalable and Generalizable Energy Time Series Forecasting

Contents & fields

This dataset contains multivariate time series from the European and international energy domain for evaluating zero-shot and covariate-aware forecasting. It covers 9 domain categories: Balancing Services, Dispatchable Generation, Non-Dispatchable Generation, Load, Market Data, Grid Data, Heat Data, Mobility / EV Charging, and Residential Load. All time series are stored in Apache Parquet format; covariate files (weather data as historical forecasts, NWP runs, and ERA5 reanalysis, plus holiday calendars) are included in sub-folders alongside each target series. All timestamps are in Coordinated Universal Time (UTC). File structure: DataClass/DatasetName-TargetName/data_file.parquet (target time series), covariates/cov_file.parquet (weather, calendar, etc.), zenodo_dataset_overview.csv (machine-readable index), and zenodo_description.html (this file). Note: Some datasets are not included due to license restrictions and must be downloaded separately from original sources.

• DataClass——Data category, e.g., Balancing Services, Dispatchable Generation
• DatasetName——Dataset name, e.g., Balancing Data Germany
• TargetName——Target variable name, e.g., Area Control Error (NRV Saldo)
• data_file.parquet——Target time series file
• cov_file.parquet——Covariate file, including weather, calendar, etc.
• zenodo_dataset_overview.csv——Machine-readable dataset index
• zenodo_description.html——Dataset description file

Research uses

This dataset is suitable for energy time series forecasting research, especially zero-shot and covariate-aware forecasting, and can be used to evaluate the generalization ability of foundation models in the energy domain.

This card was drafted from the source page; institution, coverage, time span, scale, fields and license are subject to the official page (pending human review).