Processed UKPN EPN/SPN Primary Transformer Current Forecasting Dataset

This record contains processed primary-transformer current forecasting datasets derived from UK Power Networks Open Data Portal half-hourly primary-transformer power-flow records for Eastern Power Networks (EPN) and South Eastern Power Networks (SPN). The processed arrays, preprocessing scripts, metadata, and documentation were prepared by the dataset creator. UK Power Networks provides the original source data. The files contain 12-step historical current sequences, 12-step future current targets, binary same-site adjacency matrices, and target-horizon temporal covariates. Active-power measurements are not included in the processed forecasting arrays. The processed datasets contain: - EPN: 913 primary-transformer nodes, 2023-01-01 to 2025-09-30. - SPN: 480 primary-transformer nodes, 2023-01-01 to 2025-08-31. Each dataset uses a chronological 70/10/20 train/validation/test split. Each input window contains 12 half-hourly steps, and each forecasting target contains the following 12 half-hourly steps. Source data: UK Power Networks, "Primary Transformer Power Flow Historic Half Hourly - Eastern Power Networks" and "Primary Transformer Power Flow Historic Half Hourly - South Eastern Power Networks", UK Power Networks Open Data Portal, licensed under Creative Commons Attribution 4.0 International (CC BY 4.0) according to dataset metadata. UK Power Networks is the source data publisher and is not the creator of this processed dataset. No endorsement by UK Power Networks is implied. Note on compressed file sizes The EPN and SPN files differ substantially in compressed size because the datasets are stored as compressed NumPy archives. Consecutive sliding-window samples share 11 out of 12 time steps. For SPN, the repeated block distance is 11 × 480 × 4 = 21,120 bytes, which falls within the DEFLATE history window. For EPN, the corresponding distance is 11 × 913 × 4 = 40,172 bytes, which exceeds the effective DEFLATE window. Diagnostic checks on zero values, constant windows, decimal precision, and unique values did not indicate abnormal repetition in SPN.