GOES SEP Events Dataset (1996–2024) for Deep Learning-based Flux Evolution Forecasting

Each event includes high-cadence (5-minute) proton flux measurements (≥10 MeV), soft X-ray flux, and derived features for deep learning-based SEP flux evolution forecasting.Data Content:- 174 SEP events with standardized boundary definitions (start, peak, end)- 5-minute resolution GOES proton flux (P10, ≥10 MeV) and X-ray measurements- 18-dimensional input features per time step (log-transformed flux values, temporal derivatives, and statistical features)- 24-hour prediction targets at 1-hour resolution- Event metadata: peak flux, duration, NOAA alert level (S1-S4)Format:- Individual event files (.qkl format, Python pickle)- Each file contains: event_id, event_obs_time (datetime array), event_obs_data (flux array), features (18-dim array), and metadataUsage:This dataset supports research on operational SEP forecasting, including benchmark studies of deep learning architectures (ConvLSTM, Transformer) for continuous flux evolution prediction.

The 31 test events are fixed for reproducible evaluation; training/validation events support Monte Carlo cross-validation.Data Source:Original GOES data from NOAA NCEI (ncei.noaa.gov/data/goes/). Event boundaries and quality control following standardized SEP event catalog definitions.Related Publication: